Description:FIELD OF THE INVENTION
This invention relates to Fake News Detection using a Machine Learning–Based System for Multimodal and Multilingual with Explainable and Real-Time Capabilities
BACKGROUND OF THE INVENTION
Nowadays misinformation is very common across multilingual and multimedia digital platforms and. False information leads to confusion and spreads inaccurate Data. This innovation introduces a false news detecting system that deals with wrong information and provide human-like explanations. The existing inventions have the ability to detect only English monolingual text while avoiding regional languages and code-mixed content which is widespread in India and other countries. Currents systems also fail to identify multimedia data images, videos and memes as people are engaged more and interact frequently with such content. Available models are not reliable because they lack human-like explanations which increases user trust and reduces interpretability. This invention also introduces the concept of real-time adaptability to emerging data and evolving narratives as they are not trained for dynamic updates. To conclude, this new approach introduces a new machine learning framework to analyse multilingual and multimodal content while generating explainable predictions. It not only verifies content authenticity but offers transparent explanations for each classification which builds reliability and promotes informed digital engagement.
Machine Learning models like Support Vector Machines (SVM), Naive Bayes, Logistic Regression, Random Forests, and Gradient Boosted Trees have been proved very effective for detecting fake news across digital platforms. This model extracts features from textual input like n-grams, TF-IDF scores, sentiment polarity, and lexical patterns. Classical models are computationally effective and relatively easy to understand. But they exhibit significant limitations when the input is noisy, informal, or multilingual which is very common on social media. At the moment, Ml models face certain limitations because they are mostly trained over English-based monolingual datasets. Such models perform poorly in multilingual environments like India because these are not capable of language adaptation and limited access of labeled non-English script data. Additionally, cross-lingual learning or transfer learning is restricted in traditional ML pipelines which act as a barrier and challenges scalability in developing multilingual fake news detection systems. Real world fake news include images, videos, memes, and user comments while current ML-based systems only focuses on text data. Classical ML systems do not possess an inherent ability to process and combine different types of multimedia information or diverse content sources effectively. Although researchers have combined image metadata with keyword-based tagging however proper implementation of ML-based multimodal fusion systems in existing systems have not taken place yet, which could otherwise make fake news detection more robust. A major limitation is that current ML systems are unable to provide interpretable results or clearly communicate the underlying decision-making process to end users. Whereas traditional ML models are considered more transparent than deep learning, this do not transform into real world solution when the system only provide binary outputs like “fake” or “real” without any proper explanation. Apart from this, so far this domain have not used rule-based layers, feature importance analysis, and interpretable ML frameworks (like SHAP for tree-based models), because of which it forms a trust and usability gap for non-technical users. The current models demonstrate limited capability to handle changing data in real-time. Fake news patterns, phrases, and narrative styles evolve swiftly because of ongoing current events. Traditional ML models require rare updates after training because they do not possess the ability to learn incrementally from new trends or user-flagged misinformation. The patent literature demonstrates research through sentence-level correlation methods and propagation networks but these approaches remain linked to deep learning and neural network implementations. The current available patents demonstrate few ML-specific solutions because they lack a unified approach to combine multilingual support with multimodal analysis and real-time adaptability and explainability in a lightweight ML-based system. A substantial research gap exists for creating machine learning-driven fake news detection systems which combine multilingual text processing with basic multimodal reasoning and human-readable justifications and real-time adaptability while being interpretable and computationally efficient for large-scale deployment. The proposed invention seeks to fill this gap by developing a unique ML-based pipeline which targets multilingual and media-rich misinformation environments.
US20220036011A1 A news article may include sentences and have associated comments. A embodiment determines semantic correlation between each sentence and each comment to generate correlation degrees between the sentences and the comments, determines sentence attention weights of the sentences and comment attention weights of the comments based on the correlation degrees, and detect whether the news article is fake based on latent representations of the sentences and the comments, the sentence attention weights and the comment attention weights. A list of sentences and a list of comments may be selected based on the sentence attention weights and the comment attention weights, respectively, to provide explanation for a detection result.
RESEARCH GAP: Lacks multilingual and multimodal support; limited to text and English only; does not incorporate real-time fact-aware learning.
US11494446B2 Detecting fake news involves analyzing a distribution of publishers who publish many news articles, analyzing a distribution of various topics relating to the published news articles, analyzing a social media context relating to the published news articles, and detecting fake news articles among the news articles based on the analysis of the distribution of publishers, the analysis of the distribution of the various topics, and the analysis of the social media context. Detecting fake news alternatively involves receiving online news articles including both fake online news articles and real online news articles, creating a hierarchical macro-level propagation network of the fake online news and real online news articles, the hierarchical macro-level propagation network comprising news nodes, social media post nodes, and social media repost nodes, creating a hierarchical micro-level propagation network of the fake online news and real online news articles, the hierarchical micro-level propagation network comprising reply nodes, analyzing structural and temporal features of the hierarchical macro-level propagation network, analyzing structural, temporal, and linguistic features of the hierarchical micro-level propagation network, and identifying fake news among the online news articles based on the analysis of the structural and temporal features of the hierarchical macro-level propagation network and the analysis of the structural, temporal, and linguistic features of the hierarchical micro-level propagation network.
RESEARCH GAP: Focuses on propagation networks, not on real-time detection, not multilingual or multimodal. Lacks explanation generation or user-centered transparency.
None of the prior art indicate above either alone or in combination with one another disclose what the present invention has disclosed. This invention relates to Fake News Detection using a Machine Learning–Based System for Multimodal and Multilingual with Explainable and Real-Time Capabilities
SUMMARY OF THE INVENTION
This summary is provided to introduce a selection of concepts, in a simplified format, that are further described in the detailed description of the invention.
This summary is neither intended to identify key or essential inventive concepts of the invention and nor is it intended for determining the scope of the invention.
To further clarify advantages and features of the present invention, a more particular description of the invention will be rendered by reference to specific embodiments thereof, which is illustrated in the appended drawings. It is appreciated that these drawings depict only typical embodiments of the invention and are therefore not to be considered limiting of its scope. The invention will be described and explained with additional specificity and detail with the accompanying drawings.
A machine learning-driven structure is proposed to identify and interpret misinformation covering multiple languages and information types. It’s designed to accommodate to live updates in how false information circulates, specifically on platforms like social media, where information is continuously evolving. It’s modular and that is what contributes to its originality, it doesn’t rely on third-party APIs or expert systems, and is computationally efficient for practical deployment. This facilitates scalability and optimal for today’s content rich environments, specifically in multilingual territories where news is conveyed in both local and mixed languages. Phase one of the system is, the Multilingual and Code-Mixed Preprocessing Module, engineered to interpret input in English, Hindi, or a mix of both (often detected on social media). To perform this, the framework uses a lightweight machine learning model that has been trained on labeled datasets including multilingual and code-mixed data points. Language and script detection whether Devanagari or Roman is executed and then it processes the text by applying hybrid tokenizers. After that, useful features are extracted like n-grams, TF-IDF values, sentiment polarity, aur informal language markers (like emojis or slang). These parameters enable the model to understand how people express their opinions and emotions on social media platforms. ML models such as SVMs, Logistic Regression, and Random Forests are used to categories content accurately based on the input data. The core component of the system is Multimodal Fake News Detection Engine, which not only analyses the text data but also evaluates images and its surrounding context. Text data is processed with manually engineered features and using ML models like Naive Bayes and Gradient boosted trees. If the input data is image based like memes screenshots or infographics, the system uses image processing techniques like OCR (optical character recognition) to extract the text data from the images. Along with that the system also detects visual features like colour patterns and metadata. For each media type a different kind of ML model is used and feature concatenation or similarity scoring techniques are used to combine the outputs of all these models. This detection pipeline also processes comments or replies To understand how people react on a certain content. This analysis sentiments like agree, disagree, or express strong emotions to consider it for decision-making process. This extra layer contributes to nuanced prediction by mimicking how real conversations unfold online. The system keeps itself updated through Real-Time Learning Layer. This is independent of live data from external sources, instead uses an internal database of time-tagged content that is regularly updated. The system is retrained over new examples using techniques like curriculum learning or dynamic sampling, ensuring it catches new misinformation trends as they emerge. partial_fit() function from Scikit-learn helps the model to adapt without needing to start from scratch every time. At the end, Human-Like Explanation Generator explains the user why certain content is labelled as fake or real. Unlike other AI systems this do not use complex language generation instead uses interpretable ml models. This highlights the words, phrase or image regions that influences the prediction most. With the help of tools like SHAP, Feature importance analysis, And saliency maps generate easy-to-understand visual summaries (like heatmaps), to show users what drove the model’s prediction. This leads to Transparent system and helps building trust among non-technical users.
BRIEF DESCRIPTION OF THE DRAWINGS
The illustrated embodiments of the subject matter will be understood by reference to the drawings, wherein like parts are designated by like numerals throughout. The following description is intended only by way of example, and simply illustrates certain selected embodiments of devices, systems, and methods that are consistent with the subject matter as claimed herein, wherein:
FIGURE 1 OVERALL MECHANISM
FIGURE. 2 INPUT DIAGRAM
FIGURE 3 OUTPUT DIAGRAM
The figures depict embodiments of the present subject matter for the purposes of illustration only. A person skilled in the art will easily recognize from the following description that alternative embodiments of the structures and methods illustrated herein may be employed without departing from the principles of the disclosure described herein.
DETAILED DESCRIPTION OF THE INVENTION
The detailed description of various exemplary embodiments of the disclosure is described herein with reference to the accompanying drawings. It should be noted that the embodiments are described herein in such details as to clearly communicate the disclosure. However, the amount of details provided herein is not intended to limit the anticipated variations of embodiments; on the contrary, the intention is to cover all modifications, equivalents, and alternatives falling within the scope of the present disclosure as defined by the appended claims.
It is also to be understood that various arrangements may be devised that, although not explicitly described or shown herein, embody the principles of the present disclosure. Moreover, all statements herein reciting principles, aspects, and embodiments of the present disclosure, as well as specific examples, are intended to encompass equivalents thereof.
The terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of example embodiments. As used herein, the singular forms “a",” “an” and “the” are intended to include the plural forms as well, unless the context clearly indicates otherwise. It will be further understood that the terms “comprises,” “comprising,” “includes” and/or “including,” when used herein, specify the presence of stated features, integers, steps, operations, elements and/or components, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components and/or groups thereof.
It should also be noted that in some alternative implementations, the functions/acts noted may occur out of the order noted in the figures. For example, two figures shown in succession may, in fact, be executed concurrently or may sometimes be executed in the reverse order, depending upon the functionality/acts involved.
In addition, the descriptions of "first", "second", “third”, and the like in the present invention are used for the purpose of description only, and are not to be construed as indicating or implying their relative importance or implicitly indicating the number of technical features indicated. Thus, features defining "first" and "second" may include at least one of the features, either explicitly or implicitly.
Unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as commonly understood by one of ordinary skill in the art to which example embodiments belong. It will be further understood that terms, e.g., those defined in commonly used dictionaries, should be interpreted as having a meaning that is consistent with their meaning in the context of the relevant art and will not be interpreted in an idealized or overly formal sense unless expressly so defined herein.
A machine learning-driven structure is proposed to identify and interpret misinformation covering multiple languages and information types. It’s designed to accommodate to live updates in how false information circulates, specifically on platforms like social media, where information is continuously evolving. It’s modular and that is what contributes to its originality, it doesn’t rely on third-party APIs or expert systems, and is computationally efficient for practical deployment. This facilitates scalability and optimal for today’s content rich environments, specifically in multilingual territories where news is conveyed in both local and mixed languages. Phase one of the system is, the Multilingual and Code-Mixed Preprocessing Module, engineered to interpret input in English, Hindi, or a mix of both (often detected on social media). To perform this, the framework uses a lightweight machine learning model that has been trained on labeled datasets including multilingual and code-mixed data points. Language and script detection whether Devanagari or Roman is executed and then it processes the text by applying hybrid tokenizers. After that, useful features are extracted like n-grams, TF-IDF values, sentiment polarity, aur informal language markers (like emojis or slang). These parameters enable the model to understand how people express their opinions and emotions on social media platforms. ML models such as SVMs, Logistic Regression, and Random Forests are used to categories content accurately based on the input data. The core component of the system is Multimodal Fake News Detection Engine, which not only analyses the text data but also evaluates images and its surrounding context. Text data is processed with manually engineered features and using ML models like Naive Bayes and Gradient boosted trees. If the input data is image based like memes screenshots or infographics, the system uses image processing techniques like OCR (optical character recognition) to extract the text data from the images. Along with that the system also detects visual features like colour patterns and metadata. For each media type a different kind of ML model is used and feature concatenation or similarity scoring techniques are used to combine the outputs of all these models. This detection pipeline also processes comments or replies To understand how people react on a certain content. This analysis sentiments like agree, disagree, or express strong emotions to consider it for decision-making process. This extra layer contributes to nuanced prediction by mimicking how real conversations unfold online. The system keeps itself updated through Real-Time Learning Layer. This is independent of live data from external sources, instead uses an internal database of time-tagged content that is regularly updated. The system is retrained over new examples using techniques like curriculum learning or dynamic sampling, ensuring it catches new misinformation trends as they emerge. partial_fit() function from Scikit-learn helps the model to adapt without needing to start from scratch every time. At the end, Human-Like Explanation Generator explains the user why certain content is labelled as fake or real. Unlike other AI systems this do not use complex language generation instead uses interpretable ml models. This highlights the words, phrase or image regions that influences the prediction most. With the help of tools like SHAP, Feature importance analysis, And saliency maps generate easy-to-understand visual summaries (like heatmaps), to show users what drove the model’s prediction. This leads to Transparent system and helps building trust among non-technical users.
Finally, combining all these components together generates a Powerful Machine Learning-only Solution which is very effective for Fake News Detection. This system is Multilingual, Multimodal, Transparent, and adaptable. The system avoids heavy deep learning architectures or third-party fact-checkers, making it legally compliant, easy to deploy, and highly reliable. This makes it a valuable tool for everything from monitoring media coverage and maintaining election integrity, to supporting fact-checking initiatives and combating misinformation across languages and platforms.
This pseudocode explains the core logic of machine learning-based fake news detection system. First, the system identifies the language of input data using a trained ML classifier. To handle English, Hindi, or code-mixed content, tokenization is used, which supports both Devanagari and Roman scripts. Then system extracts text-based features (such as TF-IDF) and sentiments along with visual features from images which include OCR-based text and color histograms. System also analyses the sentiment of user comments to understand the reaction. All these features are fused together into a single vector to pass them to a traditional ML classifier like SVM or Random Forest to find out whether the content is fake, partially true, or real. At last, the system uses SHAP or similar interpretable techniques to highlight the most influential features behind the forecast, providing an explainable and trustworthy output to the user.
Algorithm:
Input: User content (Text, Image, Comments)
Output: Veracity label (Fake, Partially True, Real) + Explanation

# 1. Preprocessing
lang = detect_language(text) # ML classifier: Hindi / English / Code-Mixed
tokenized_text = hybrid_tokenizer(text, lang) # Roman + Devanagari
text_features = extract_text_features(tokenized_text) # TF-IDF, n-grams, sentiment

# 2. Image Feature Extraction (if image is present)
if image:
ocr_text = run_ocr(image)
image_features = extract_visual_features(image) # color hist, metadata, OCR

# 3. Comment Sentiment Analysis
comment_features = []
for comment in comments:
sentiment = analyze_sentiment(comment) # lexicon/supervised
comment_features.append(sentiment)

# 4. Feature Fusion
combined_features = concatenate(text_features, image_features, comment_features)

# 5. Classification
veracity_label = classifier.predict(combined_features) # SVM / RandomForest / GBT

# 6. Explanation Generation
important_features = shap_explainer(classifier, combined_features)
highlighted_tokens = visualize_influential_words(important_features)

# 7. Output
return veracity_label, highlighted_tokens
ADVANTAGES OF THE INVENTION:
• It can detect English, Hindi, and code-mixed Hinglish using machine learning models trained on diverse linguistic data.
• Textual and visual data (text + image) is handled efficiently. It is also extendable to audio/video analysis in future iterations.
• Generates clear, machine learning–based explanations using attention scores and interpretable feature attribution methods (e.g., SHAP/LIME visualizations).
• Periodic model fine-tuning using time-stamped datasets and emerging misinformation patterns to simulate real-time updates.
• Optimized, lightweight models (e.g., distilled BERT, MobileBERT, TinyCNN) suitable for deployment on mobile and low-power edge devices.
• Incorporates crowd-sourced feedback to iteratively improve model confidence and performance using semi-supervised learning strategies.
• Detects manipulation in visual media (e.g., images contradicting textual claims) through joint representation learning.
• Achieves higher accuracy, precision, recall, and F1 scores while maintaining transparency and improving public trust through explainability.  
, Claims:1. A machine learning–based system for fake news detection, comprising:
a multilingual preprocessing module configured to process text in English, Hindi, and code-mixed Hinglish using a language identification classifier and hybrid tokenizers capable of handling Roman and Devanagari scripts;
a multimodal detection engine configured to analyze textual, visual, and comment-based inputs; and
a classification module configured to determine whether the input is fake, partially true, or real.
2. The system as claimed in claim 1, wherein the multilingual preprocessing module extracts linguistic features including n-grams, TF-IDF values, sentiment polarity, stylistic cues, and informal markers such as emojis and slang.
3. The system as claimed in claim 1, wherein the multimodal detection engine applies optical character recognition (OCR) on images to extract embedded text, and further derives visual features including color histograms and image metadata.
4. The system as claimed in claim 1, wherein the multimodal detection engine analyzes comment sentiment using lexicon-based or supervised machine learning classifiers to determine audience reaction features such as agreement, disagreement, or emotional intensity.
5. The system as claimed in claim 1, wherein the classification module employs traditional machine learning algorithms selected from Support Vector Machines (SVM), Random Forest, Logistic Regression, and Gradient Boosted Trees.
6. The system as claimed in claim 1, further comprising a real-time learning layer configured to incrementally update the classification model using new misinformation trends, time-tagged datasets, and crowd-sourced feedback, without retraining from scratch.
7. The system as claimed in claim 6, wherein said real-time learning layer uses incremental learning techniques including partial fit, curriculum learning, or dynamic sampling to adapt to evolving misinformation patterns.
8. The system as claimed in claim 1, further comprising an explainability module configured to generate human-understandable justifications for predictions by highlighting influential features using SHAP, feature importance analysis, permutation importance, and saliency mapping.
9. The system as claimed in claim 8, wherein said explainability module presents outputs in the form of highlighted text regions, heatmaps, or feature summaries to increase transparency and user trust.
10. The system as claimed in claim 1, wherein the system is optimized for deployment on mobile and low-power devices by using lightweight machine learning models including distilled BERT, MobileBERT, or TinyCNN for efficient real-time operation.