1. A computer-implemented sentiment analysis system (100) for emotion-aware opinion mining, the system (100) comprising: a user interface (102), wherein the user interface (102) is hosted on a user device (104), wherein the user interface (102) is configured to capture a plurality of structured and/or unstructured review data stream in textual and/or audio format from a plurality of sources; a backend processing assembly (106) bilaterally connected to the user interface (102), the backend processing assembly (106) further comprising: a preprocessing module (108) configured to: receive the unstructured/structured textual input; process and convert the received audio input into text; and normalize linguistic content; an emotion annotator (110) receiving input from the preprocessing module (108), and the emotion annotator (110) configured to automatically generate a plurality of multi-dimensional emotional annotations; an emotion-guided polarized capsule dropout (EG-PCD) engine (112) operationally coupled to the emotion annotator (110), the emotion-guided polarized capsule dropout (EG-PCD) engine (112) configured to perform emotion-aware feature selection and adaptive regularization, wherein the emotion-guided polarized capsule dropout (EG-PCD) engine (112) groups the emotion-aware features into a plurality of capsules representing a higher-level semantic–emotional concept, wherein the emotion-guided polarized capsule dropout (EG-PCD) engine (112) dynamically calculates dropout probabilities for each of the capsules; a bidirectional long short-term memory (BiLSTM) sequence modeling engine (114) operatively coupled to the emotion-guided polarized capsule dropout (EG-PCD) engine (112), the bidirectional long short-term memory (BiLSTM) sequence modeling engine (114) configured to process the capsules in forward and backward temporal directions to preserve contextual dependencies and capture long-range emotional continuity; and a sentiment inference engine (116) operationally coupled to the bidirectional long short-term memory (BiLSTM) sequence modeling engine (114), the sentiment inference engine (116) configured to perform emotion-weighted feature fusion, wherein the sentiment inference engine (116) includes: a plurality of neural layers (118); and a regression layer (120), wherein the plurality of neural layers (118), and the regression layer (120) performs accurate sentiment classification using sentiment labelling and continuous sentiment scoring.

2. The system (100) as claimed in claim 1, wherein the multi-dimensional emotional annotations include, emotion category, emotional polarity, and emotion intensity.

3. The system (100) as claimed in claim 2, wherein the multi-dimensional emotional annotations generated by the emotion annotator (110) act as guiding metadata for the emotion-guided polarized capsule dropout (EG-PCD) engine (112).

4. The system (100) as claimed in claim 1, wherein the emotion-guided polarized capsule dropout (EG-PCD) engine (112) calculates an emotion score to dynamically adjust the dropout probability for each of the capsule.

5. The system (100) as claimed in claim 4, wherein the emotion-guided polarized capsule dropout (EG-PCD) engine (112) drops the capsule with low emotional score is dropped with higher dropout probability and the capsule with high emotional score is dropped with lower dropout probability.

6. The system (100) as claimed in claim 1, wherein the bidirectional long short-term memory (BiLSTM) sequence modeling engine (114) enables accurate handling of complex emotions containing, contrastive opinions, sarcasm, and/or mixed sentiments.

7. The system (100) as claimed in claim 1, wherein the emotion-weighted feature fusion vis the sentiment inference engine (116) is performed by merging semantic meaning, emotional cues, and contextual dependencies.

8. A method (200) for emotion-guided sentiment analysis, the method (200) comprising: capturing a plurality of structured and/or unstructured review data stream in textual and/or audio format from a plurality of sources via a user interface (102), wherein the user interface (102) is hosted on a user device (104); preprocessing structured and/or unstructured review data stream via a preprocessing module (108) hosted on a backend processing assembly (106) to generate normalized linguistic content; generating emotion annotations comprising emotion type, emotional polarity, and emotion intensity via an emotion annotator (110); extracting and grouping the emotion-aware features into a plurality of capsules representing a higher-level semantic–emotional concept via an emotion-guided polarized capsule dropout (EG-PCD) engine (112); dynamically applying capsule dropout by regulating capsule activation and dropout probabilities according to emotional relevance; processing retained capsule representations using a bidirectional long short-term memory (BiLSTM) sequence modeling engine (114) to model bidirectional contextual and emotional dependencies; and generating sentiment predictions using a sentiment inference engine (116).

9. The method (200) as claimed in claim 8, wherein each of the capsules associate with a dropout probability that is inversely proportional to the emotion scores.

10. The method (200) as claimed in claim 8, wherein the bidirectional long short-term memory (BiLSTM) sequence modeling engine (114) captures bidirectional contextual dependencies while preserving emotional relevance, improving comprehension of complex linguistic patterns.

Description:FIELD OF DISCLOSURE
[0001] The present disclosure generally relates to a sentiment analysis system and specifically relates to a sentiment analysis system for emotion-aware opinion mining
BACKGROUND OF THE DISCLOSURE
[0002] Embodiments of the present invention generally relate to a system and method for sentiment analysis through emotion-guided polarized capsule dropout integrated with bidirectional long short-term memory (BiLSTM) networks.
[0003] Sentiment analysis, also known as opinion mining, aims to automatically identify, extract, and classify subjective information from textual data, such as emotions, attitudes, and opinions expressed by users. With the rapid growth of social media platforms, e-commerce websites, review forums, and feedback systems, the volume of unstructured textual data has increased exponentially. This has made sentiment analysis an essential tool for businesses, policymakers, and researchers seeking insights into public opinion, customer satisfaction, and decision-making trends. Despite significant progress in sentiment analysis over the past decade, accurately interpreting the semantic and contextual nuances embedded in user-generated reviews remains a persistent and complex challenge.
[0004] Additionally, the inherently complex nature of natural language presents substantial obstacles that limit the effectiveness and robustness of existing sentiment analysis approaches. One of the fundamental challenges lies in the semantic ambiguity of natural language. Words and phrases often carry multiple meanings depending on context, domain, or usage patterns. For example, a term such as “unpredictable” may convey negative sentiment in the context of product reliability but may be interpreted positively when describing a movie plot or a sports match.
[0005] Traditional lexicon-based and rule-based sentiment analysis methods typically rely on predefined sentiment dictionaries, which assign fixed polarity scores to words. Such approaches fail to dynamically adapt to context, resulting in inaccurate sentiment classification when words shift polarity across domains. As a result, such systems struggle to capture nuanced opinions, especially in heterogeneous datasets spanning multiple topics or industries.
[0006] Another significant limitation of existing sentiment analysis systems is the inability to effectively handle linguistic complexity, including sarcasm, irony, and idiomatic expressions. Sarcasm often involves expressing negative sentiment using seemingly positive language, such as “Great, another software update that broke everything.” Most conventional machine learning and deep learning models interpret sentiment based on surface-level word polarity and frequency, leading to misclassification in sarcastic or ironic statements. Similarly, idiomatic expressions and colloquial phrases convey sentiment implicitly and are highly culture- and language-specific. Without explicit modelling of such expressions, existing systems often fail to correctly infer the intended emotional tone, thereby reducing overall prediction accuracy.
[0007] The hierarchical structure of language further complicates sentiment analysis. As natural language is not merely a flat sequence of words but is organized into phrases, clauses, and sentences that interact hierarchically to convey meaning. Early machine learning approaches, such as bag-of-words and n-gram models, ignore word order and syntactic structure, leading to a loss of crucial contextual information. Although more advanced deep learning models like recurrent neural networks (RNNs) and long short-term memory (LSTM) networks capture sequential dependencies, they still exhibit limitations in representing hierarchical relationships across different levels of linguistic abstraction. This shortcoming becomes particularly evident in long and complex reviews where sentiment may shift across sentences or depend on higher-level discourse structures.
[0008] Closely related to hierarchical modelling is the challenge of capturing spatial and relational dependencies between words. Sentiment is often expressed through interactions between multiple words rather than individual tokens. For instance, negations (“not good”), intensifiers (“very bad”), and contrastive conjunctions (“although”, “but”) significantly alter sentiment polarity. Many existing models process text sequentially and focus primarily on local dependencies, making it difficult to capture long-range interactions and relation/al patterns. Further, such solutions still require large amounts of labelled data and substantial computational resources, limiting their practicality in low-resource or real-time applications.
[0009] Another limitation is that many sentiment analysis solutions predominantly focus on sequential dependencies, assuming that sentiment information flows linearly through text. While sequential modelling is valuable, it does not fully represent the multi-dimensional nature of language, where sentiment cues may arise from non-adjacent words or implicit relationships across sentences. For example, a user review may begin with neutral or positive statements and conclude with a critical remark that dominates the overall sentiment. Sequential models may dilute or misinterpret such patterns, particularly when sentiment-bearing words are sparsely distributed or contextually dependent on earlier discourse.
[0010] Furthermore, emotional richness and intensity are often inadequately represented in existing solutions. Most sentiment analysis systems classify text into coarse-grained categories such as positive, negative, or neutral, overlooking the spectrum of human emotions and their varying intensities. Even when multi-class sentiment labels are used, models frequently fail to distinguish closely related sentiments, such as frustration versus disappointment or excitement versus satisfaction. This limitation reduces the interpretability and usefulness of sentiment analysis outputs, especially in applications that require fine-grained emotional insights.
[0011] Another critical limitation involves the robustness and generalizability of existing sentiment analysis models. Many deep learning-based approaches achieve high accuracy on benchmark datasets but perform poorly when applied to real-world data with domain shifts, noisy language, spelling errors, code-mixed text, or informal grammar. User-generated content often includes emojis, abbreviations, misspellings, and mixed languages, which are not adequately handled by models trained on clean, curated datasets. This lack of robustness undermines the reliability of sentiment analysis systems in practical deployment scenarios.
[0012] Thus, while existing sentiment analysis techniques have achieved notable success, they remain constrained by several fundamental limitations. Most current approaches emphasize sequential modelling while neglecting deeper inter-word and contextual relationships that are critical for accurate sentiment understanding. Therefore, there is a need for more advanced sentiment analysis frameworks that integrate emotional awareness, contextual sensitivity, and robust representation learning to effectively capture the complexity of human language and opinions.
[0013] Therefore, there disclosed invention presents more advanced and accurate sentiment analysis system for emotion-aware opinion mining.
SUMMARY
[0014] The following is a summary description of illustrative embodiments of the invention. It is provided as a preface to assist those skilled in the art to more rapidly assimilate the detailed design discussion which ensues and is not intended in any way to limit the scope of the claims which are appended hereto in order to particularly point out the invention.
[0015] Embodiments in accordance with the present invention provide a computer-implemented sentiment analysis system for emotion-aware opinion mining. Embodiments in accordance with the present invention also provide a method for emotion-guided sentiment analysis.
[0016] Embodiments of the present invention may provide a number of advantages depending on its particular configuration. First, embodiments of the present application provide a computer-implemented sentiment analysis system (100) for emotion-aware opinion mining. Next, embodiments of the present application provide a method for emotion-guided sentiment analysis.
[0017] The present disclosure solves all the major limitation of traditional system.
[0018] An objective of the present disclosure is to provide an advanced sentiment analysis system capable of accurately identifying and classifying sentiments.
[0019] Another objective of the present disclosure is to overcome limitations of conventional sentiment analysis techniques by enabling emotion-guided feature learning that captures nuanced emotional expressions, mixed emotions, and sentiment shifts within data.
[0020] Another objective of the present disclosure is to selectively retain emotionally significant features while suppressing emotionally irrelevant or noisy features, thereby improving robustness and generalization.
[0021] Another objective of the present disclosure is to capture bidirectional and long-range contextual dependencies while preserving emotional continuity across sequences.
[0022] Another objective of the present disclosure is to reduce overfitting commonly associated with deep learning models.
[0023] Yet another objective of the present disclosure is to eliminate dependence on handcrafted or manually engineered features by automatically learning semantic, emotional, and contextual representations from data.
[0024] Yet another objective of the present disclosure is to enable reliable sentiment analysis of emotionally ambiguous, sarcastic, contrastive, or mixed-emotion reviews.
[0025] Yet another objective of the present disclosure is to provide a scalable and domain-independent sentiment analysis framework adaptable to diverse applications.
[0026] In the light of above disclosure, in an aspect of the present disclosure a computer-implemented sentiment analysis system for emotion-aware opinion mining is disclosed herein. The system comprising a user interface. The user interface hosted on a user device. The user interface configured to capture a plurality of structured and/or unstructured review data stream in textual and/or audio format from a plurality of sources. The backend processing assembly bilaterally connected to the user interface, and the backend processing assembly further comprising a preprocessing module configured to receive the unstructured/structured textual input. The backend processing assembly further comprising process and convert the received audio input into text. The backend processing assembly further comprising normalize linguistic content. The backend processing assembly further comprising an emotion annotator receiving input from the preprocessing module, and the emotion annotator configured to automatically generate a plurality of multi-dimensional emotional annotations. The backend processing assembly further comprising an emotion-guided polarized capsule dropout (EG-PCD) engine operationally coupled to the emotion annotator, and the emotion-guided polarized capsule dropout (EG-PCD) engine configured to perform emotion-aware feature selection and adaptive regularization. The emotion-guided polarized capsule dropout (EG-PCD) engine groups the emotion-aware features into a plurality of capsules representing a higher-level semantic–emotional concept. The emotion-guided polarized capsule dropout (EG-PCD) engine dynamically calculates dropout probabilities for each of the capsules. The backend processing assembly further comprising a bidirectional long short-term memory (BiLSTM) sequence modeling engine operatively coupled to the emotion-guided polarized capsule dropout (EG-PCD) engine, and the bidirectional long short-term memory (BiLSTM) sequence modeling engine configured to process the capsules in forward and backward temporal directions to preserve contextual dependencies and capture long-range emotional continuity. The backend processing assembly further comprising a sentiment inference engine operationally coupled to the bidirectional long short-term memory (BiLSTM) sequence modeling engine, and the sentiment inference engine configured to perform emotion-weighted feature fusion. The sentiment inference engine includes a plurality of neural layers, and a regression layer. The plurality of neural layers, and the regression layer performs accurate sentiment classification using sentiment labelling and continuous sentiment scoring.
[0027] In one embodiment, the multi-dimensional emotional annotations include, emotion category, emotional polarity, and emotion intensity.
[0028] In one embodiment, the multi-dimensional emotional annotations generated by the emotion annotator act as guiding metadata for the emotion-guided polarized capsule dropout (EG-PCD) engine.
[0029] In one embodiment, the emotion-guided polarized capsule dropout (EG-PCD) engine calculates an emotion score to dynamically adjust the dropout probability for each of the capsule.
[0030] In one embodiment, the emotion-guided polarized capsule dropout (EG-PCD) engine drops the capsule with low emotional score is dropped with higher dropout probability and the capsule with high emotional score is dropped with lower dropout probability.
[0031] In one embodiment, the bidirectional long short-term memory (BiLSTM) sequence modeling engine enables accurate handling of complex emotions containing, contrastive opinions, sarcasm, and/or mixed sentiments.
[0032] In one embodiment, the emotion-weighted feature fusion vis the sentiment inference engine is performed by merging semantic meaning, emotional cues, and contextual dependencies.
[0033] In another aspect of the present disclosure, a method for emotion-guided sentiment analysis is disclosed herein. The method comprising capturing a plurality of structured and/or unstructured review data stream in textual and/or audio format from a plurality of sources via a user interface. The user interface is hosted on a user device. The method also comprising preprocessing structured and/or unstructured review data stream via a preprocessing module hosted on a backend processing assembly to generate normalized linguistic content. The method also comprising generating emotion annotations comprising emotion type, emotional polarity, and emotion intensity via an emotion annotator. The method also comprising extracting and grouping the emotion-aware features into a plurality of capsules representing a higher-level semantic–emotional concept via an emotion-guided polarized capsule dropout (EG-PCD) engine. The method also comprising dynamically applying capsule dropout by regulating capsule activation and dropout probabilities according to emotional relevance. The method also comprising processing retained capsule representations using a bidirectional long short-term memory (BiLSTM) sequence modeling engine to model bidirectional contextual and emotional dependencies. The method also comprising generating sentiment predictions using a sentiment inference engine.
[0034] In one embodiment, each of the capsules associate with a dropout probability that is inversely proportional to the emotion intensity
[0035] In one embodiment, the bidirectional long short-term memory (BiLSTM) sequence modeling engine capture bidirectional contextual dependencies while preserving emotional relevance, improving comprehension of complex linguistic patterns.
[0036] These and other advantages will be apparent from the present application of the embodiments and solves abovementioned limitations in the traditional system.
[0037] The preceding is a simplified summary to provide an understanding of some embodiments of the present invention. This summary is neither an extensive nor exhaustive overview of the present invention and its various embodiments. The summary presents selected concepts of the embodiments of the present invention in a simplified form as an introduction to the more detailed description presented below. As will be appreciated, other embodiments of the present invention are possible utilizing, alone or in combination, one or more of the features set forth above or described in detail below.
[0038] These elements, together with the other aspects of the present disclosure and various features are pointed out with particularity in the claims annexed hereto and form a part of the present disclosure. For a better understanding of the present disclosure, its operating advantages, and the specified object attained by its uses, reference should be made to the accompanying drawings and descriptive matter in which there are illustrated exemplary embodiments of the present disclosure.
BRIEF DESCRIPTION OF THE DRAWINGS
[0039] The above and still further features and advantages of embodiments of the present invention will become apparent upon consideration of the following detailed description of embodiments thereof, especially when taken in conjunction with the accompanying drawings, and wherein:
[0040] FIG. 1 illustrates a block diagram of a computer-implemented sentiment analysis system for emotion-aware opinion mining, according to an embodiment of the present invention; and
[0041] FIG. 2 illustrates a flowchart of a method for emotion-guided sentiment analysis, according to an embodiment of the present invention.
[0042] The headings used herein are for organizational purposes only and are not meant to be used to limit the scope of the description or the claims. As used throughout this application, the word "may" is used in a permissive sense (i.e., meaning having the potential to), rather than the mandatory sense (i.e., meaning must). Similarly, the words “include”, “including”, and “includes” mean including but not limited to. To facilitate understanding, like reference numerals have been used, where possible, to designate like elements common to the figures. Optional portions of the figures may be illustrated using dashed or dotted lines, unless the context of usage indicates otherwise.
DETAILED DESCRIPTION
[0043] The following description includes the preferred best mode of one embodiment of the present invention. It will be clear from this description of the invention that the invention is not limited to these illustrated embodiments but that the invention also includes a variety of modifications and embodiments thereto. Therefore, the present description should be seen as illustrative and not limiting. While the invention is susceptible to various modifications and alternative constructions, it should be understood, that there is no intention to limit the invention to the specific form disclosed, but, on the contrary, the invention is to cover all modifications, alternative constructions, and equivalents falling within the spirit and scope of the invention as defined in the claims.
[0044] In any embodiment described herein, the open-ended terms "comprising," "comprises,” and the like (which are synonymous with "including," "having” and "characterized by") may be replaced by the respective partially closed phrases "consisting essentially of," consists essentially of," and the like or the respective closed phrases "consisting of," "consists of, the like.
[0045] As used herein, the singular forms “a”, “an”, and “the” designate both the singular and the plural, unless expressly stated to designate the singular only.
[0046] FIG. 1 illustrates a block diagram of a computer-implemented sentiment analysis system 100 for emotion-aware opinion mining, according to an embodiment of the present invention.
[0047] The system 100 may be comprising a user interface 102, a backend processing assembly 106, a preprocessing module 108, an emotion annotator 110, an emotion-guided polarized capsule dropout (EG-PCD) engine 112, a bidirectional long short-term memory (BiLSTM) sequence modeling engine 114, and a sentiment inference engine 116.
[0048] The user interface 102 may be hosted on a user device 104. The user interface 102 may be configured to capture a plurality of structured and/or unstructured review data stream in textual and/or audio format from a plurality of sources.
[0049] In an embodiment of the present disclosure, the user interface 102 may be hosted on any user device (104) including, but not limited to, smartphones, tablets, laptop computers, desktop systems, or wearable computing devices. The user interface (102) may be implemented as a web-based interface, a mobile application, or a desktop application. The user interface (102) may enable a user to collect review content, view sentiment analysis results, and manage review sources.
[0050] In an embodiment of the present disclosure, the structured data may include predefined rating fields, survey responses, or feedback forms, while the unstructured data may include free-form text reviews, social media posts, voice feedback, or conversational transcripts. The user interface 102 may support real-time and batch data submission. In one implementation, the user interface 102 may support textual input capture using standard text entry components such as, but not limited to, text boxes, comment fields, chat interfaces, or document upload mechanisms. In an embodiment of the present disclosure, the textual data may be obtained directly from the users or ingested via application programming interfaces (APIs) connected to external platforms such as, but not limited to, e-commerce websites, social media platforms, customer support systems, or survey tools.
[0051] In one embodiment, the user interface 102 may support multi-source data aggregation, enabling the ingestion of review data from multiple platforms simultaneously. Such platforms may include e-commerce marketplaces, social networking services, review portals, call-centre feedback systems, or enterprise customer relationship management (CRM) systems.
[0052] In an embodiment of the present disclosure, the user interface 102 may support audio input capture using microphones integrated into the user device (104). Captured audio data is processed using any automatic speech recognition (ASR) techniques to convert speech into textual form prior to sentiment analysis. The automatic speech recognition (ASR) functionality may be implemented using on-device speech processing engines or cloud-based speech-to-text services.
[0053] In an embodiment of the present disclosure, the user interface 102 may be implemented using web technologies including, but not limited to, HTML, CSS, JavaScript, and frontend frameworks such as React, Angular, or Vue.js, or mobile development frameworks including, but not limited to, Android SDK, iOS SDK, Flutter, or React Native. In one embodiment, the user interface 102 may be connected with a memory unit or cloud storage infrastructure. In an embodiment of the present disclosure, the user interface 102 may employ encryption protocols such as, but not limited to, HTTPS, TLS, or equivalent secure communication standards to ensure data integrity and confidentiality.
[0054] In some embodiments, the user interface 102 may include administrative and configuration controls allowing the user/s to configure data sources, define input formats, select analysis modes, and visualize sentiment analysis outputs through dashboards, charts, or reports.
[0055] The backend processing assembly 106 may be bilaterally connected to the user interface 102 and the backend processing assembly 106 further comprising the preprocessing module 108, the emotion annotator 110, the emotion-guided polarized capsule dropout (EG-PCD) engine 112, the bidirectional long short-term memory (BiLSTM) sequence modeling engine 114, and the sentiment inference engine 116.
[0056] In an embodiment of the present disclosure, the bilateral communication between the user interface 102 and the backend processing assembly 106 may be facilitated using RESTful APIs, WebSocket connections, or message-based communication protocols. In an embodiment of the present disclosure, the backend processing assembly 106 may be any microcontroller or a microprocessor or any remote cloud-based processing infrastructure. In one embodiment, the backend processing assembly 106 may be deployed on cloud-based infrastructure, on-premise servers, or edge computing platforms, to manage concurrent connections and processing workloads.
[0057] In one embodiment, the backend processing assembly 106 may be bilaterally connected to the user interface 102 through a wireless communication network enabling bidirectional data exchange between the user interface 102 and the backend processing assembly 106. The bilateral connectivity may allow the backend processing assembly 106 to receive review data streams from the user interface 102 and to transmit sentiment analysis results, status updates, and system feedback back to the user interface 102 in real time or near real time.
[0058] In some embodiment, the bilateral connection maybe established using client–server architecture, wherein the user interface 102 may operate as a client application and the backend processing assembly 106 may operate as a centralized processing server. Communication between the client and server may be facilitated through RESTful application programming interfaces (APIs), GraphQL interfaces, or remote procedure call (RPC) mechanisms. In some embodiment, the bilateral connectivity may support real-time communication using persistent communication channels such as WebSocket connections, server-sent events (SSE), or message queuing protocols.
[0059] The preprocessing module 108 may be configured to receive the unstructured/structured textual input. The preprocessing module 108 may be configured to process and convert the received audio input into text. The preprocessing module 108 may be configured to normalize linguistic content.
[0060] In an embodiment of the present disclosure, the preprocessing module 108 may be configured to receive structured and/or unstructured textual input from the user interface (102), extracted from the external data sources.
In some embodiments, the preprocessing module 108 may further be configured to receive audio input and convert the audio input into corresponding textual data prior to downstream analysis. The audio-to-text conversion may be performed using automatic speech recognition (ASR) techniques implemented either on-device or through cloud-based speech-to-text services. The audio-to-text conversion enables the system 100 to process voice-based user feedback in a manner consistent with text-based reviews.
[0061] In one embodiment, the preprocessing module 108 may perform linguistic normalization on the received textual data to ensure uniformity and consistency. Linguistic normalization may include operations such as, but not limited to, tokenization, lowercasing, punctuation handling, spelling correction, expansion of contractions, removal of extraneous symbols, and normalization of whitespace.
In some embodiments, the preprocessing module 108 may be configured to perform language detection and filtering, enabling multilingual support and selection of appropriate language-specific processing pipelines. In some embodiments, the unsupported languages may be flagged or routed to auxiliary processing modules.
[0062] In an embodiment of the present disclosure, the preprocessing module 108 may additionally perform noise reduction and content filtering, including removal of stopwords, handling of emojis and emoticons, normalization of slang or colloquial expressions, and elimination of irrelevant metadata that may adversely affect sentiment analysis performance. In an embodiment of the present disclosure, the preprocessing module 108 may be implemented using natural language processing libraries and frameworks including, but not limited to, tokenizers, lemmatizers, or stemming algorithms.
[0063] In an embodiment of the present disclosure, the preprocessing module 108 may transform heterogeneous raw input data into a standardized, machine-readable textual representation, enabling accurate downstream emotion-aware feature extraction and sentiment analysis. In some embodiments, the preprocessing module 108 may include validation mechanisms to filter, normalize, and format incoming review data. Such mechanisms may include language detection, noise filtering, punctuation normalization, encoding standardization, and verification of input completeness prior to transmission to downstream processing modules.
[0064] The emotion annotator 110 may be receiving input from the preprocessing module 108, and the emotion annotator 110 configured to automatically generate a plurality of multi-dimensional emotional annotations.
[0065] The multi-dimensional emotional annotations may include, emotion category, emotional polarity, and emotion intensity.
[0066] In an embodiment of the present disclosure, the emotion annotator 110 may be configured to receive preprocessed textual data from the preprocessing module 108. The emotion annotator 110 may automatically analyse the received textual data to generate a plurality of multi-dimensional emotional annotations corresponding to linguistic units including words, phrases, sentences, or document-level segments.
[0067] In one embodiment, the multi-dimensional emotional annotations may be generated by the emotion annotator 110 include at least an emotion category, an emotional polarity, and an emotion intensity score. The emotion category may represent discrete emotional states such as, but not limited to, joy, anger, sadness, fear, disgust, or surprise. Emotional polarity may indicate positive, negative, or neutral emotional orientation, while emotion intensity quantifies the strength or magnitude of the expressed emotion.
[0068] In one implementation, the emotion annotator 110 may employ one or more emotion recognition techniques including, but not limited to, emotion lexicon-based analysis, machine learning classifiers, deep neural networks, or hybrid emotion detection models. Emotion annotations may be generated at multiple granularities including, but not limited to, token-level, phrase-level, sentence-level, or review-level representations.
[0069] In some embodiment, the emotion annotator 110 may normalize and aggregate emotion signals across linguistic units to generate consistent emotion vectors with preservation of emotional continuity across textual sequences.
[0070] In an embodiment of the present disclosure, the emotion annotator 110 may be implemented as a standalone software component, a microservice, or an embedded neural processing module within the backend processing assembly 106. In one embodiment, the emotion annotator 110 may be implemented using natural language processing (NLP) and machine learning technologies. The emotion annotator 110 may utilize token-level and sentence-level embedding models to convert textual input into vector representations capturing semantic and affective information. Such embeddings may be generated using word-level, subword-level, or contextual embedding techniques.
[0071] In an embodiment of the present disclosure, the emotion annotator 110 may employ deep learning-based emotion classification models, including but not limited to convolutional neural networks (CNNs), recurrent neural networks (RNNs), Long Short-Term Memory (LSTM) networks, Bidirectional LSTM networks, transformer-based architectures, or combinations thereof. These models may be trained on labeled emotion datasets to identify discrete emotion categories.
[0072] In an embodiment of the present disclosure, the emotion annotator 110 may incorporate emotion lexicon-based techniques, wherein predefined emotion dictionaries or affective lexicons may be used to map words or phrases to emotion categories and polarity values. Lexicon-derived emotion scores may be combined with neural model outputs to improve robustness and coverage. The emotion annotator 110 may apply emotional polarity detection algorithms to classify emotional orientation as positive, negative, or neutral. Polarity detection may be performed using supervised classifiers, probabilistic models, or neural scoring mechanisms.
[0073] In one implementation, the emotion annotator 110 may compute emotion intensity values by aggregating confidence scores, activation magnitudes, or attention weights produced by the underlying emotion recognition models. Emotion intensity values may be normalized to a predefined scale to ensure compatibility with downstream processing modules. In some embodiments, outputs from multiple emotion detection techniques I ncluding lexicon-based scoring, neural classification, and contextual analysis—may be fused using weighted averaging, ensemble learning, or rule-based arbitration to generate multi-dimensional emotional annotations.
[0074] In some embodiments, the emotion annotator 110 may be implemented using programming frameworks and libraries for natural language processing and deep learning, and may execute on general-purpose processors, graphics processing units (GPUs), tensor processing units (TPUs), or equivalent hardware accelerators. The emotion annotator 110 may be deployed on cloud infrastructure, on-premise servers, or edge computing platforms.
[0075] The emotion-guided polarized capsule dropout (EG-PCD) engine 112 may be operationally coupled to the emotion annotator 110, and the emotion-guided polarized capsule dropout (EG-PCD) engine 112 configured to perform emotion-aware feature selection and adaptive regularization.
[0076] The multi-dimensional emotional annotations may be generated by the emotion annotator 110 act as guiding metadata for the emotion-guided polarized capsule dropout (EG-PCD) engine 112.
[0077] The emotion-guided polarized capsule dropout (EG-PCD) engine 112 may group the emotion-aware features into a plurality of capsules representing a higher-level semantic–emotional concept.
[0078] The emotion-guided polarized capsule dropout (EG-PCD) engine 112 may dynamically calculate dropout probabilities for each of the capsules.
[0079] The emotion-guided polarized capsule dropout (EG-PCD) engine 112 may calculate an emotion score to dynamically adjust the dropout probability for each of the capsule.
[0080] The emotion-guided polarized capsule dropout (EG-PCD) engine 112 may drop the capsule with low emotional score is dropped with higher dropout probability and the capsule with high emotional score is dropped with lower dropout probability.
[0081] In an embodiment of the present disclosure, the emotion-guided polarized capsule dropout (EG-PCD) engine 112 may receive multi-dimensional emotional annotations from the emotion annotator 110. The emotion-guided polarized capsule dropout (EG-PCD) engine 112 may perform emotion-aware feature selection and adaptive regularization by integrating emotional polarity and emotion intensity directly into the feature learning process.
[0082] In an embodiment of the present disclosure, the guiding metadata may be utilized to modulate capsule formation, activation strength, and dropout behaviour, thereby ensuring that emotionally salient information is preferentially preserved during training. In an embodiment of the present disclosure, each of the capsule may encode a higher-level semantic–emotional concept derived from combinations of linguistic, contextual, and emotional attributes. The capsule representations may correspond to phrases, clauses, opinion targets, or sentiment-bearing expressions within the textual data.
[0083] In one implementation, the emotion-guided polarized capsule dropout (EG-PCD) engine 112 may dynamically calculate capsule-specific dropout probabilities based on emotional polarity, emotion intensity, and capsule relevance scores. In an exemplary embodiment, the capsules associated with high emotional intensity or strong emotional polarity may be assigned lower dropout probabilities, while capsules associated with low emotional relevance may be assigned higher dropout probabilities, thereby enabling adaptive, emotion-regulated regularization. The emotion-guided polarized capsule dropout (EG-PCD) engine 112 may selectively preserve emotionally meaningful capsule representations while suppressing noisy or less relevant features, resulting in improved model generalization, stability, and interpretability compared to uniform or random dropout techniques.
[0084] In an embodiment of the present disclosure, the emotion-guided polarized capsule dropout (EG-PCD) engine 112 may be implemented using a combination of artificial intelligence, deep learning, and natural language processing technologies, executed on one or more computing devices forming part of the backend processing assembly 106. In one embodiment, the emotion-guided polarized capsule dropout (EG-PCD) engine 112 may be realized as a software-implemented neural processing module configured to execute within a deep learning framework capable of supporting custom neural layers and adaptive regularization mechanisms.
[0085] In one embodiment, the emotion-guided polarized capsule dropout (EG-PCD) engine 112 may be implemented using deep learning libraries and frameworks that support neural network construction, training, and inference. Such frameworks enable the creation of capsule-based feature representations, dynamic dropout operations, and gradient-based optimization. The emotion-guided polarized capsule dropout (EG-PCD) engine 112 may utilize tensor-based computation engines to efficiently process high-dimensional semantic–emotional feature vectors.
[0086] In one implementation, the capsule representations within the emotion-guided polarized capsule dropout (EG-PCD) engine 112 may be realized using capsule network architectures, wherein groups of neurons form vector-based capsules that encode higher-level semantic–emotional concepts. Capsule routing mechanisms may be employed to preserve relationships between features, and capsule activation values may be modulated using emotion-derived metadata received from the emotion annotator 110.
[0087] In some embodiments, emotion-aware feature integration technologies may be used to incorporate emotional annotations into the capsule representations. Emotional polarity and emotion intensity values may be combined with semantic feature vectors through feature concatenation, weighted scaling, or gating functions, enabling emotional information to directly influence capsule activation and dropout behavior. The adaptive dropout functionality of the emotion-guided polarized capsule dropout (EG-PCD) engine 112 may be implemented using custom regularization layers that extend conventional dropout operations. The capsule-specific dropout probabilities may be dynamically computed using mathematical or rule-based functions driven by emotion scores, rather than being uniformly or randomly assigned.
[0088] In one embodiment, numerical computation and optimization within the emotion-guided polarized capsule dropout (EG-PCD) engine 112 may be supported by linear algebra libraries and automatic differentiation engines, enabling efficient backpropagation and training stability. Optimization algorithms may be employed to iteratively update model parameters while maintaining emotion-aware regularization.
[0089] In some embodiments, the emotion-guided polarized capsule dropout (EG-PCD) engine 112 may be executed on general-purpose computing hardware or hardware accelerators, including central processing units (CPUs), graphics processing units (GPUs), tensor processing units (TPUs), or equivalent accelerators to enable scalable and real-time processing of large volumes of review data.
[0090] In some embodiment, the emotion-guided polarized capsule dropout (EG-PCD) engine 112 may be deployed using cloud-based infrastructure, containerized environments, or microservice-based architectures, allowing flexible scaling and seamless integration with other components of the system 100. Communication with the emotion annotator 110 may be achieved through shared memory, tensor exchange, or service-based interfaces.
[0091] The bidirectional long short-term memory (BiLSTM) sequence modeling engine 114 may be operatively coupled to the emotion-guided polarized capsule dropout (EG-PCD) engine 112, and the bidirectional long short-term memory (BiLSTM) sequence modeling engine 114 configured to process the capsules in forward and backward temporal directions to preserve contextual dependencies and capture long-range emotional continuity.
[0092] The bidirectional long short-term memory (BiLSTM) sequence modeling engine 114 may enable accurate handling of complex emotions containing, contrastive opinions, sarcasm, and/or mixed sentiments.
[0093] In an embodiment of the present disclosure, the bidirectional long short-term memory (BiLSTM) sequence modeling engine 114 may employ a pair of LSTM networks arranged in opposite directions, wherein one LSTM processes the capsule sequence from a beginning of the input text toward an end, and the other LSTM processes the capsule sequence from the end toward the beginning. The outputs from the forward and backward LSTM networks may be concatenated, summed, or otherwise fused to generate a context-enriched capsule sequence representation.
[0094] In one implementation, the bidirectional long short-term memory (BiLSTM) sequence modeling engine 114 may incorporate emotion-preserving gating mechanisms, wherein emotional relevance embedded within the capsule representations influences LSTM gate activations. This enables the bidirectional long short-term memory (BiLSTM) sequence modeling engine 114 to maintain emotional context across long textual spans and resolve sentiment transitions within complex reviews.
[0095] In one embodiment, the bidirectional long short-term memory (BiLSTM) sequence modeling engine 114 may enable accurate handling of contrastive opinions, sarcastic expressions, and mixed sentiments by modelling both preceding and succeeding contextual cues. The bidirectional long short-term memory (BiLSTM) sequence modeling engine 114 may detect sentiment reversals, negations, and emotionally conflicting phrases that may not be captured by unidirectional sequence models.
[0096] In an embodiment of the present disclosure, the bidirectional long short-term memory (BiLSTM) sequence modeling engine 114 may be implemented using deep learning frameworks capable of supporting recurrent neural network architectures. The bidirectional long short-term memory (BiLSTM) sequence modeling engine 114 may utilize optimized LSTM cell implementations that support gradient-based learning, backpropagation through time, and efficient sequence processing.
[0097] In some alternative embodiments, the bidirectional long short-term memory (BiLSTM) sequence modeling engine 114 may be replaced or augmented with stacked BiLSTM layers, gated recurrent unit (GRU) layers, or hybrid recurrent architectures, depending on application-specific performance requirements. In some embodiments, the bidirectional long short-term memory (BiLSTM) sequence modeling engine 114 may be combined with temporal attention mechanisms applied over the BiLSTM outputs, allowing the system 100 to selectively emphasize emotionally significant capsule sequences while maintaining bidirectional contextual understanding. In an embodiment of the present disclosure, the bidirectional long short-term memory (BiLSTM) sequence modeling engine 114 may execute on general-purpose processors or hardware accelerators, including CPUs, GPUs, TPUs, or equivalent computing hardware.
[0098] The sentiment inference engine 116 may be operationally coupled to the bidirectional long short-term memory (BiLSTM) sequence modeling engine 114, and the sentiment inference engine 116 configured to perform emotion-weighted feature fusion.
[0099] The sentiment inference engine 116 may include a plurality of neural layers 118, and a regression layer 120.
[0100] The plurality of neural layers 118, and the regression layer 120 may perform accurate sentiment classification using sentiment labelling and continuous sentiment scoring.
[0101] The emotion-weighted feature fusion via the sentiment inference engine 116 may be performed by merging semantic meaning, emotional cues, and contextual dependencies.
[0102] In an embodiment of the present disclosure, the sentiment inference engine 116 may perform emotion-weighted feature fusion, enabling joint consideration of semantic, emotional, and contextual information for sentiment prediction. In an embodiment of the present disclosure, the sentiment inference engine 116 may include a plurality of neural layers 118 arranged as one or more fully connected layers, followed by a regression layer 120 or classification layer. The plurality of neural layers 118 may be configured to progressively transform the fused feature representations into a sentiment-discriminative feature space.
[0103] In one embodiment, the plurality of neural layers 118 and the regression layer 120 perform accurate sentiment classification by generating either discrete sentiment labels, including positive, negative, or neutral classes, or continuous sentiment scores representing sentiment intensity along a predefined scale. The regression layer 120 may employ activation functions appropriate for classification or scoring tasks.
[0104] In one embodiment, emotion-weighted feature fusion performed by the sentiment inference engine 116 may include merging semantic meaning, emotional cues, and contextual dependencies using weighted combination techniques. In one embodiment, the sentiment inference engine 116 may apply normalization, regularization, or confidence calibration techniques to improve prediction stability and interpretability. Feature importance scores or contribution weights may be generated to provide insight into how emotional and contextual features affect sentiment outcomes.
[0105] In an embodiment of the present disclosure, the sentiment inference engine 116 may be implemented using deep learning frameworks capable of supporting multilayer neural networks, regression functions, and gradient-based optimization. The sentiment inference engine 116 may execute on general-purpose processors or hardware accelerators, enabling scalable training and real-time inference.
[0106] FIG. 2 illustrates a flowchart of a method 200 for emotion-guided sentiment analysis, according to an embodiment of the present invention.
[0107] The method 200 may comprise the following steps.
[0108] At 202, capturing a plurality of structured and/or unstructured review data stream in textual and/or audio format from a plurality of sources via a user interface 102.
[0109] The user interface 102 may be hosted on a user device 104.
[0110] At 204, preprocessing structured and/or unstructured review data stream via a preprocessing module 108 hosted on a backend processing assembly 106 to generate normalized linguistic content.
[0111] At 206, generating emotion annotations comprising emotion type, emotional polarity, and emotion intensity via an emotion annotator 110.
[0112] At 208, extracting and grouping the emotion-aware features into a plurality of capsules representing a higher-level semantic–emotional concept via an emotion-guided polarized capsule dropout (EG-PCD) engine 112.
[0113] At 210, dynamically applying capsule dropout by regulating capsule activation and dropout probabilities according to emotional relevance;
[0114] At 212, processing retained capsule representations using a bidirectional long short-term memory (BiLSTM) sequence modeling engine 114 to model bidirectional contextual and emotional dependencies.
[0115] At 214, generating sentiment predictions using a sentiment inference engine 116.
[0116] Each of the capsules may associate with a dropout probability that is inversely proportional to the emotion scores.
[0117] The bidirectional long short-term memory (BiLSTM) sequence modeling engine 114 may capture bidirectional contextual dependencies while preserving emotional relevance, improving comprehension of complex linguistic patterns.
[0118] In one embodiment, the preprocessing module 108 may perform linguistic normalization operations such as, but not limited to, tokenization, noise removal, lemmatization, stop-word filtering, and syntactic normalization, thereby generating normalized linguistic content suitable for downstream emotional and semantic analysis. The normalized linguistic content may be provided to an emotion annotator 110 configured to automatically generate emotion annotations. The emotion annotations comprise at least an emotion type, emotional polarity, and emotion intensity, thereby enabling multi-dimensional emotional representation of the review data stream.
[0119] In one embodiment, the emotion-annotated linguistic content may be processed by an emotion-guided polarized capsule dropout (EG-PCD) engine 112 to extract emotion-aware features and group them into capsules, corresponding to emotionally salient opinion patterns. In one embodiment, the emotion-guided polarized capsule dropout (EG-PCD) engine 112 calculates dropout probabilities according to emotional score. The retained capsules with lower dropout possibilities may be processed using the bidirectional long short-term memory (BiLSTM) sequence modelling engine 114 to model bidirectional contextual dependencies and emotional continuity across the review data stream.
[0120] In one embodiment, the sentiment inference engine 116 may generate sentiment predictions, including sentiment labels and/or continuous sentiment scores, based on fused semantic, emotional, and contextual information. The output of the sentiment inference engine 116 may be transmitted back to the user interface 102 in a user-friendly manner. In some embodiments, the output of the sentiment inference engine 116 may be in report format or any other pre-determined or user defined format.
[0121] The following table 1 depicts evaluation of EG-PCD-BiLSTM using different K-fold cross validation values.
Table 1. Evaluation of EG-PCD-BiLSTM using different K-fold cross validation values
K-fold values Datasets Accuracy (%) Precision (%) Recall (%) F1-score (%) Specificity (%)
K=2 Amazon Reviews for Sentiment Analysis 96.42 96.10 95.88 95.99 96.35
IMDB 97.61 97.52 97.48 97.45 97.55
K=3 Amazon Reviews for Sentiment Analysis 97.10 96.88 96.8 96.84 96.9
IMDB 97.61 97.52 97.48 97.45 97.55
K=5 Amazon Reviews for Sentiment Analysis 97.94
97.94
97.94
97.94
97.94

IMDB 99.03
99.05
99.03
99.03
99.03

K=6 Amazon Reviews for Sentiment Analysis 97.3 97.2 97.1 97.15 97.25
IMDB 98.45 98.3 98.25 98.27 98.35
K=7 Amazon Reviews for Sentiment Analysis 96.88 96.70 96.55 96.62 96.85
IMDB 98.2 98 97.9 97.95 98.15
[0122] The disclosed invention offers several advantages. The disclosed invention provides a substantial technical advancement in the field of sentiment analysis by enabling emotion-sensitive sentiment classification that goes beyond conventional polarity detection. By explicitly modeling emotion category, emotional polarity, and emotion intensity, the system 100 achieves a deeper understanding of subjective language, particularly in reviews expressing mixed, conflicting, or subtle emotional states. The disclosed invention provides more accurate and contextually aligned sentiment predictions compared to existing sentiment analysis systems.
[0123] A key advantage of the disclosed invention lies in its emotion-guided learning framework, wherein emotional annotations actively guide feature selection and regularization. Unlike traditional machine learning models that depend on manually engineered features, and deep learning models that implicitly learn representations without emotional awareness, the system 100 automatically extracts and prioritizes emotionally salient semantic features. This reduces feature noise and enhances the relevance of learned representations, leading to improved classification stability and robustness.
[0124] The integration of Emotion-Guided Polarized Capsule Dropout (EG-PCD) mechanism offers superior regularization mechanism. Conventional dropout techniques randomly suppress neurons or features without regard to emotional relevance, often resulting in loss of important sentiment cues. In contrast, the disclosed invention dynamically adjusts capsule dropout probabilities based on emotion scores, ensuring that emotionally significant capsules are retained while emotionally weak or irrelevant capsules are selectively dropped. This emotion-adaptive dropout mechanism improves generalization performance and prevents overfitting.
[0125] Another advantage of the disclosed invention is its ability to retain long-range emotional and contextual dependencies. Traditional deep learning models, including standard RNNs and CNN-based architectures, struggle to preserve long-term dependencies and emotional continuity. The disclosed invention uses the BiLSTM-based sequence modeling engine 114 to process the capsule representations in both forward and backward directions, enabling accurate interpretation of complex linguistic constructs such as sarcasm, contrastive opinions, and emotionally evolving narratives.
[0126] The disclosed invention further overcomes limitations of attention-based sentiment models, which often emphasize syntactic importance or word-level relevance while overlooking emotional polarity and intensity. By embedding emotional guidance directly into capsule formation and dropout regulation, the disclosed invention ensures that emotional polarity is not merely inferred post-hoc but is intrinsically embedded within the learned representations. This leads to more robust sentiment inference, particularly in emotionally ambiguous or nuanced text.
[0127] Unlike conventional capsule networks that primarily focus on spatial or hierarchical feature relationships, the disclosed invention introduces emotion-sensitive capsule activation, linking capsule relevance to emotional polarity and intensity. The capsule representations enhance semantic–emotional contextual learning, enabling the system 100 to distinguish between structurally similar but emotionally divergent expressions, a task where existing capsule-based approaches underperform.
[0128] The disclosed system also provides improved interpretability, as capsule activations and emotion-weighted dropout probabilities offers transparency, which is particularly beneficial for enterprise, regulatory, and decision-support applications where explainability of outputs is critical. Additionally, the disclosed invention exhibits scalability and domain adaptability, allowing it to be applied across diverse application domains such as product reviews, social media analysis, customer feedback systems, healthcare narratives, and financial opinion mining. By jointly addressing emotional awareness, contextual dependency preservation, feature retention, and generalization, the invention achieves superior accuracy, robustness, and reliability compared to existing sentiment analysis techniques.
[0129] While the invention has been described in connection with what is presently considered to be the most practical and various embodiments, it is to be understood that the invention is not to be limited to the disclosed embodiments, but on the contrary, is intended to cover various modifications and equivalent arrangements included within the spirit and scope of the appended claims.
[0130] This written description uses examples to disclose the invention, including the best mode, and also to enable any person skilled in the art to practice the invention, including making and using any devices or systems and performing any incorporated methods. The patentable scope the invention is defined in the claims, and may include other examples that occur to those skilled in the art. Such other examples are intended to be within the scope of the claims if they have structural elements that do not differ from the literal language of the claims, or if they include equivalent structural elements within substantial differences from the literal languages of the claims.
, Claims:I/We Claim:
1. A computer-implemented sentiment analysis system (100) for emotion-aware opinion mining, the system (100) comprising:
a user interface (102),
wherein the user interface (102) is hosted on a user device (104),
wherein the user interface (102) is configured to capture a plurality of structured and/or unstructured review data stream in textual and/or audio format from a plurality of sources;
a backend processing assembly (106) bilaterally connected to the user interface (102), the backend processing assembly (106) further comprising:
a preprocessing module (108) configured to:
receive the unstructured/structured textual input;
process and convert the received audio input into text; and
normalize linguistic content;
an emotion annotator (110) receiving input from the preprocessing module (108), and the emotion annotator (110) configured to automatically generate a plurality of multi-dimensional emotional annotations;
an emotion-guided polarized capsule dropout (EG-PCD) engine (112) operationally coupled to the emotion annotator (110), the emotion-guided polarized capsule dropout (EG-PCD) engine (112) configured to perform emotion-aware feature selection and adaptive regularization,
wherein the emotion-guided polarized capsule dropout (EG-PCD) engine (112) groups the emotion-aware features into a plurality of capsules representing a higher-level semantic–emotional concept,
wherein the emotion-guided polarized capsule dropout (EG-PCD) engine (112) dynamically calculates dropout probabilities for each of the capsules;
a bidirectional long short-term memory (BiLSTM) sequence modeling engine (114) operatively coupled to the emotion-guided polarized capsule dropout (EG-PCD) engine (112), the bidirectional long short-term memory (BiLSTM) sequence modeling engine (114) configured to process the capsules in forward and backward temporal directions to preserve contextual dependencies and capture long-range emotional continuity; and
a sentiment inference engine (116) operationally coupled to the bidirectional long short-term memory (BiLSTM) sequence modeling engine (114), the sentiment inference engine (116) configured to perform emotion-weighted feature fusion,
wherein the sentiment inference engine (116) includes:
a plurality of neural layers (118); and
a regression layer (120),
wherein the plurality of neural layers (118), and the regression layer (120) performs accurate sentiment classification using sentiment labelling and continuous sentiment scoring.
2. The system (100) as claimed in claim 1, wherein the multi-dimensional emotional annotations include, emotion category, emotional polarity, and emotion intensity.
3. The system (100) as claimed in claim 2, wherein the multi-dimensional emotional annotations generated by the emotion annotator (110) act as guiding metadata for the emotion-guided polarized capsule dropout (EG-PCD) engine (112).
4. The system (100) as claimed in claim 1, wherein the emotion-guided polarized capsule dropout (EG-PCD) engine (112) calculates an emotion score to dynamically adjust the dropout probability for each of the capsule.
5. The system (100) as claimed in claim 4, wherein the emotion-guided polarized capsule dropout (EG-PCD) engine (112) drops the capsule with low emotional score is dropped with higher dropout probability and the capsule with high emotional score is dropped with lower dropout probability.
6. The system (100) as claimed in claim 1, wherein the bidirectional long short-term memory (BiLSTM) sequence modeling engine (114) enables accurate handling of complex emotions containing, contrastive opinions, sarcasm, and/or mixed sentiments.
7. The system (100) as claimed in claim 1, wherein the emotion-weighted feature fusion vis the sentiment inference engine (116) is performed by merging semantic meaning, emotional cues, and contextual dependencies.
8. A method (200) for emotion-guided sentiment analysis, the method (200) comprising:
capturing a plurality of structured and/or unstructured review data stream in textual and/or audio format from a plurality of sources via a user interface (102),
wherein the user interface (102) is hosted on a user device (104);
preprocessing structured and/or unstructured review data stream via a preprocessing module (108) hosted on a backend processing assembly (106) to generate normalized linguistic content;
generating emotion annotations comprising emotion type, emotional polarity, and emotion intensity via an emotion annotator (110);
extracting and grouping the emotion-aware features into a plurality of capsules representing a higher-level semantic–emotional concept via an emotion-guided polarized capsule dropout (EG-PCD) engine (112);
dynamically applying capsule dropout by regulating capsule activation and dropout probabilities according to emotional relevance;
processing retained capsule representations using a bidirectional long short-term memory (BiLSTM) sequence modeling engine (114) to model bidirectional contextual and emotional dependencies; and
generating sentiment predictions using a sentiment inference engine (116).
9. The method (200) as claimed in claim 8, wherein each of the capsules associate with a dropout probability that is inversely proportional to the emotion scores.
10. The method (200) as claimed in claim 8, wherein the bidirectional long short-term memory (BiLSTM) sequence modeling engine (114) captures bidirectional contextual dependencies while preserving emotional relevance, improving comprehension of complex linguistic patterns.