Description:FORM 2

THE PATENTS ACT, 1970
(39 of 1970)
&
THE PATENT RULES, 2003

COMPLETE SPECIFICATION
(See Section 10 and Rule 13)

Title of invention:

METHOD AND SYSTEM FOR REINFORCEMENT LEARNING BASED ESTIMATION OF FINANCIAL FIDELITY

Applicant

Tata Consultancy Services Limited
A company Incorporated in India under the Companies Act, 1956
Having address:
Nirmal Building, 9th floor,
Nariman point, Mumbai 400021,
Maharashtra, India

Preamble to the description:

The following specification particularly describes the invention and the manner in which it is to be performed.
TECHNICAL FIELD
[001] The disclosure herein generally relates to field of reinforcement learning based sentiment analysis, and, more particularly, reinforcement learning based sentiment analysis to gain financial fidelity of a person, or an entity based on data gathered through mass-communication platforms.

BACKGROUND
[002] Financial institution such as banks, insurance agencies are frequently engaged in assessing the credit risk of counterparties on an ongoing basis, i.e., the risk that a party cannot service its debt obligations. For this purpose, the financial institutions utilize credit scores issued by credit information bureaus like CIBIL (Credit Information Bureau India Limited), Equifax, Experian, and the like. The credit score is an important indicator for lenders to assess the ability of a customer to repay loans. It is prepared based on data gathered from lenders and is consolidated in the Credit Information Report or CIR. Financial institutions use your credit history and score as one of the factors to determine whether to lend any money, how much and at what rate.
[003] In early days, credit decisions were made by bank credit officials; these officials knew the applicant, since they usually lived in the same town, and would make credit decisions based on this knowledge. In the 1970's, the FICO score made credit far more available, effectively removing the credit officer from the process. However, the risk management function still needs to be done. Lenders, such as banks and credit card companies, use credit scores to evaluate the potential risk posed by lending money to consumers. With the time, computers became an indispensable part of credit scoring and transformed the traditional way of assessing the credit through computers ability of performing multivariate-multifunctional analysis. The traditional credit scoring transformations do not work well for groups of variables, especially when data is partially or completely missing. Additionally, as a consequence of the need for human quality control, traditional transformations are also limited in the amount of data which can be reasonably processed. Each transformation and filling-in operation may require a human to invest a significant amount of time to analyze one or more data fields, and then carefully manipulate the contents of the field. Such restraints limit the number of fields to an amount which can be understood by a single person in a reasonable period of time, and, as a result, there are relatively few risk models (such as a FICO score by Fair Isaac Corporation, Experian bureau scores, Pinnacle by Equifax, or Precision by TransUnion) with more than a few tens of variables (e.g. a FICO score is based on five basic metrics, including payment history, credit utilization, length of credit history, types of credit used, and recent searches for credit).
[004] However, there are instances wherein credit score may not be readily available, especially in case of new business entity, or a very small entity (such as micro-enterprises). In such cases the financial institution requires a substitute source to gain confidence about financial fidelity of the customer. Moreover, decision-makers assessing credit risk, it is desirable to supplement quantitative risk measures with qualitative information. It is very challenging for the financial institutions to assess the customer for making a lending base while providing financial support in terms of loan or credits.

SUMMARY
[005] Embodiments of the present disclosure present technological improvements as solutions to one or more of the above-mentioned technical problems recognized by the inventors in conventional systems. For example, in one embodiment, a method of estimating financial fidelity of the entity through sentiment sensitivity index is provided. The method includes, receiving, via one or more hardware processors, a plurality of news items from a plurality of sources wherein the plurality of sources comprises newswires, e-newspapers, e-magazines, news blogs, news broadcasts, social media feeds and news feeds on the mass-communication platform. The system 100 captures relevant news in real-time from news feed and wire feed especially focusing on the news items related to corporate actions, regulatory actions, political context, business context or from technological context. The presence of the news on any mass-communication platform about an entity under analysis that can be captured from web is taken into consideration to prepare as input for sentiment analysis model. The method further includes, categorizing, via the one or more hardware processors, the plurality of news items by parsing the plurality of news items from a plurality of pre-set business rules. The pre-set business rules are made to identify all such news items which about the entity surfacing on the web with respect to one or more parameter that may be important for a lender/ financial institution to know while considering providing credits/ loan to the entity. The method further includes, via the one or more hardware processors, feeding each category of the news items to a pre-trained AI/ML model. For training the AI/ML model, all the identified key words are utilized in preparing a labelled dataset. The labelled dataset is the key requirement to train the AI/ML model. The method further includes, via the one or more hardware processors, deriving, category-wise sentiment score of the plurality of news items. The AI/ML model classifies the news items input as labeled dataset of relevant keywords into categories. The categories are based on topics which are common among the plurality of news items and group the news items in one bucket. The basis of categorization is to recognize the fact that various types of news impact sentiment in different ways and to a different degree. Sentiment scores are derived for each category. The natural language processing (NLP) and artificial intelligence (AI) tools are used for sentiment score generation and further polarity determination is subjected to reinforced learning to improve the scoring. Unsupervised learning is used to finetune the categorization and polarity accuracy based on user’s feedback. The method further includes, specifying, via the one or more hardware processors, a look-back period for the plurality of news items in each category. The look back period is specified, for which news should be looked back for determining as on date sentiment. Selection of a look-back period has a high impact on deriving more realistic sentiment score. The method further includes, via the one or more hardware processors, applying a time decay factor to each news item by using a decay function wherein the time decay factor represents degraded sentiments for news items farthest in time. A decay sentiment score is obtained by multiplying the category-based sentiment score of the news items with time decay factor to determine how much score to back to add on date for a news which has occurred in past. The method further includes, via the one or more hardware processors, adjusting, the category-wise sentiment score of each news item by factoring the specified look-back period and the time decay factor. Time decay factor and look-back period are relational. The farthest look-back period will have high decay factor and the nearest look-back period will have lowest decay factor. The method further includes, via the one or more hardware processors, assigning weightage to each category of the news items and multiplying the weightage with the category-wise sentiment score. The method further includes, via the one or more hardware processors, deriving entity-level sentiment score by consolidating the category-wise sentiment score of the plurality of news items and normalize the consolidated sentiment score to sense the business reputation of the entity. The negative value of sentiment score for an entity is considered as Market Insight Pricing Index (MIPI) score for that entity. The financial fidelity of the entity is represented by sentiment sensitivity index (SSI).
[006] In another aspect, a system for estimating financial fidelity of the entity using sentiment sensitivity index is provided. The system includes at least one memory storing programmed instructions; one or more Input /Output (I/O) interfaces; and one or more hardware processors, a sentiment analysis module, operatively coupled to a corresponding at least one memory, wherein the system is configured to receive, a plurality of news items from a plurality of sources wherein the plurality of sources comprises newswires, e-newspapers, e-magazines, news blogs, news broadcasts, social media feeds and news feeds on the mass-communication platforms. The system 100 captures relevant news in real-time from news feed and wire feed especially focusing on the news items related to corporate actions, regulatory actions, political context, business context or from technological context. The presence of the news on any mass-communication platform about an entity under analysis that can be captured from web is taken into consideration to prepare as input for sentiment analysis model. Further, the system is configured to categorize, the plurality of news items by parsing the plurality of news items from a plurality of pre-set business rules. The pre-set business rules are made to identify all such news items which about the entity surfacing on the web with respect to one or more parameter that may be important for a lender/ financial institution to know while considering providing credits/ loan to the entity. Further, the system is configured to feed each category of the news items to a pre-trained AI/ML model. For training the AI/ML model, all the identified key words are utilized in preparing a labelled dataset. The labelled dataset is the key requirement to train the AI/ML model. Further, the system is configured to derive, category-wise sentiment score of the plurality of news items. The AI/ML model classifies the news items input as labeled dataset of relevant keywords into categories. The categories are based on topics which are common among the plurality of news items and group the news items in one bucket. The basis of categorization is to recognize the fact that various types of news impact sentiment in different ways and to a different degree. Sentiment score is derived for each category. The natural language processing (NLP) and artificial intelligence (AI) tools are used for sentiment score generation and further polarity determination is subjected to reinforced learning to improve the scoring. Unsupervised learning is used to finetune the categorization and polarity accuracy based on user’s feedback. Further, the system is configured to specify a look-back period for the plurality of news items in each category. The look back period is specified, for which news should be looked back for determining as on date sentiment. Selection of a look-back period has a high impact on deriving more realistic sentiment score. Further, the system is configured to apply a time decay factor to each news item by using a decay function wherein the time decay factor represents degraded sentiments for news items farthest in time. A decay sentiment score is obtained by multiplying the category-based sentiment score of the news items with time decay factor to determine how much score to back to add on date for a news which has occurred in past. Further, the system is configured to adjust the category-wise sentiment score of each news item by factoring the specified look-back period and the time decay factor. Time decay factor and look-back period are relational. The farthest look-back period will have high decay factor and the nearest look-back period will have lowest decay factor. Further, the system is configured to assign weightage to each category of the news items and multiplying the weightage with the category-wise sentiment score. Further, the system is configured to derive entity-level sentiment score by consolidating the category-wise sentiment score of the plurality of news items and normalize the consolidated sentiment score to sense the business reputation of the entity. The negative value of sentiment score for an entity is considered as Market Insight Pricing Index (MIPI) score for that entity. The financial fidelity of the entity is represented by sentiment sensitivity index (SSI).
[007] In yet another aspect, a computer program product including a non-transitory computer-readable medium having embodied therein a computer program for estimating financial fidelity of the entity using sentiment sensitivity index is provided. The computer readable program, when executed on a computing device, causes the computing device to receive, via sentiment analysis module, executed by the one or more hardware processors, a plurality of news items from a plurality of sources wherein the plurality of sources comprises newswires, e-newspapers, e-magazines, news blogs, news broadcasts, social media feeds and news feeds on the mass-communication platforms.
[008] The system 100 captures relevant news in real-time from news feed and wire feed especially focusing on the news items related to corporate actions, regulatory actions, political context, business context or from technological context. The presence of the news on any mass-communication platform about an entity under analysis that can be captured from web is taken into consideration to prepare as input for sentiment analysis model. The computer readable program, when executed on a computing device, causes the computing device to categorize, via the one or more hardware processors, the plurality of news items by parsing the plurality of news items from a plurality of pre-set business rules. The pre-set business rules are made to identify all such news items which about the entity surfacing on the web with respect to one or more parameter that may be important for a lender/ financial institution to know while considering providing credits/ loan to the entity. The computer readable program, when executed on a computing device, causes the computing device to feed, via the one or more hardware processors, each category of the news items to a pre-trained AI/ML model. For training the AI/ML model, all the identified key words are utilized in preparing a labelled dataset. The labelled dataset is the key requirement to train the AI/ML model. The computer readable program, when executed on a computing device, causes the computing device to derive, via the one or more hardware processors, category-wise sentiment score of the plurality of news items. The AI/ML model classifies the news items input as labeled dataset of relevant keywords into categories. The categories are based on topics which are common among the plurality of news items and group the news items in one bucket. The basis of categorization is to recognize the fact that various types of news impact sentiment in different ways and to a different degree. Sentiment scores are derived for each category. The natural language processing (NLP) and artificial intelligence (AI) tools are used for sentiment score generation and further polarity determination is subjected to reinforced learning to improve the scoring. Unsupervised learning is used to finetune the categorization and polarity accuracy based on user’s feedback. The computer readable program, when executed on a computing device, causes the computing device to specify, via the one or more hardware processors, a look-back period for the plurality of news items in each category. The look back period is specified, for which news should be looked back for determining as on date sentiment. Selection of a look-back period has a high impact on deriving more realistic sentiment score. The computer readable program, when executed on a computing device, causes the computing device to apply, via the one or more hardware processors, a time decay factor to each news item by using a decay function wherein the time decay factor represents degraded sentiments for news items farthest in time. A decay sentiment score is obtained by multiplying the category-based sentiment score of the news items with time decay factor to determine how much score to back to add on date for a news which has occurred in past. The computer readable program, when executed on a computing device, causes the computing device to adjust, via the one or more hardware processors, the category-wise sentiment score of each news item by factoring the specified look-back period and the time decay factor. Time decay factor and look-back period are relational. The farthest look-back period will have high decay factor and the nearest look-back period will have lowest decay factor. The computer readable program, when executed on a computing device, causes the computing device to assign, via the one or more hardware processors, weightage to each category of the news items and multiplying the weightage with the category-wise sentiment score. The computer readable program, when executed on a computing device, causes the computing device to derive, via the one or more hardware processors, entity-level sentiment score by consolidating the category-wise sentiment score of the plurality of news items and normalize the consolidated sentiment score to sense the business reputation of the entity. The negative value of sentiment score for an entity is considered as Market Insight Pricing Index (MIPI) score for that entity. The financial fidelity of the entity is represented by sentiment sensitivity index (SSI).
[009] It is to be understood that both the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the invention, as claimed.

BRIEF DESCRIPTION OF THE DRAWINGS
[010] The accompanying drawings, which are incorporated in and constitute a part of this disclosure, illustrate exemplary embodiments and, together with the description, serve to explain the disclosed principles:
[011] FIG. 1 illustrates a system for sensing the sensitivity index of an entity according to some embodiments of the present disclosure.
[012] FIG. 2 is a functional block diagram depicting reinforcement learning involved in deriving sentiment score according to some embodiments of the present disclosure.
[013] FIG. 3 illustrates a method of sensing the sensitivity index in accordance with some embodiments of the present disclosure.
[014] FIG. 4 is a block diagram depicting model training for performing sentiment analysis in accordance with some embodiments of the present disclosure.
[015] FIG. 5 is a user interface representation of the news analysis for deriving sentiment sensitivity index (SSI) in accordance with some embodiments of the present disclosure.

DETAILED DESCRIPTION OF EMBODIMENTS
[016] Exemplary embodiments are described with reference to the accompanying drawings. In the figures, the left-most digit(s) of a reference number identifies the figure in which the reference number first appears. Wherever convenient, the same reference numbers are used throughout the drawings to refer to the same or like parts. While examples and features of disclosed principles are described herein, modifications, adaptations, and other implementations are possible without departing from the scope of the disclosed embodiments.
[017] As used herein, the term “sentiment sensitivity index” refers to a mass communication platform index of sentiments about a customer (i.e. an entity or an organization) wherein the sentiments are utilized to gather intelligence on the customer for analyzing dynamic credit profile of the customer through sentiments.
[018] In the present disclosure, a method of assessing financial fidelity of a customer or an entity is provided wherein the assessment of the financial fidelity is based on mining, analyzing, and interpreting raw data gathered from newswire, social media, financial reports, capital market and research agencies. Currently there is no system or method which provides real-time intelligence on the customer using input from newswire, social media, financial reports, capital market and research agencies. The input sources are considered as unstructured or semi-structured data sources. In the finance context, unstructured or semi-structured data sources, such as newswires, company reports, financial reports, capital market analysis, research agency findings, quarterly conference calls, financial news, social media, and analyst reports are open for analysis and can provide meaningful insights. Data-driven decision-making is a key concept for identifying, supporting, and justifying credit decisions taken by the credit agencies in the absence of the credit score or as an additional validation tool that supplements the credit score with another instrument that defines financial fidelity of the customer. Such justifications by applying sentiment and topic analysis to the unstructured and semi-structured data gathered can offer additional insights, as seasoned analysts use them to disseminate in-depth research to experienced investors.
[019] There are no existing technical solutions that can provide customer business credibility to financial institutions such as banks in the absence of credit history in the MSME market. The present disclosure bridges the gap where credit history of the customer is unavailable to the bank and bank has to provide lending base for judgement. The unique sentiment sensitivity index (SSI) calculation provides the bank with an insight and scale in taking a decision about the customer. The SSI provides for the customer specific trade finance related position of the entity based on sentiment sensitivity index derived by analysing a plurality of instances about the entity surfacing into mass communication platforms. The sentiment sensitivity index considers social, political, financial, economic, environmental and trade related aspects associated with the entity to understand its reputation in the market. Accordingly, banks choose to finance and strategize their pricing for product and services. This index can be used to gather intelligence on the customer to gain an insight into its ever-changing credit profile. Predictive insights from SSI can help make up for absence of credit history of the customer by generating a score based on its dynamic lendability assessment. SSI is a near real time intelligence on the customer which uses inputs from newswires, social media, financial reports, capital market, research agency and rating agency feeds. This market insight index is then used in a multifactor pricing model to generate the final index. The present disclosure suggests that even when established credit risk indicators and financial news are considered, the sentiment and a subset of topics are correlated with changes in the credit default, indicating a fundamental relationship between quantitative risk metric and analyst reports.
[020] Referring now to the drawings, and more particularly to FIG. 1 through FIG. 5, where similar reference characters denote corresponding features consistently throughout the figures, there are shown preferred embodiments, and these embodiments are described in the context of the following exemplary system and/or method.
[021] FIG. 1 illustrates an exemplary block diagram of a system 100 for sensing sentiment sensitivity index (or SSI) of an entity according to some embodiments of the present disclosure.
[022] In an embodiment, the system 100 includes one or more processors 104, communication interface device(s) or input/output (I/O) interface(s) 106, and one or more data storage devices or memory 102 operatively coupled to the one or more processors 104. The one or more processors 104 that are hardware processors can be implemented as one or more microprocessors, microcomputers, microcontrollers, digital signal processors, central processing units, state machines, graphics controllers, logic circuitries, and/or any devices that manipulate signals based on operational instructions. Among other capabilities, the processor(s) are configured to fetch and execute computer-readable instructions stored in the memory. In the context of the present disclosure, the expressions ‘processors’ and ‘hardware processors’ may be used interchangeably. In an embodiment, the system 100 can be implemented in a variety of mobile computing systems, such as mobile devices, laptop computers, notebooks, hand-held devices, workstations, mainframe computers, servers, a network cloud, and the like. The I/O interface (s) 106 may include a variety of software and hardware interfaces, for example, a web interface, a graphical user interface, and the like and can facilitate multiple communications within a wide variety of networks and protocol types, including wired networks, for example, LAN, cable, etc., and wireless networks, such as WLAN, cellular, or satellite. In an embodiment, the I/O interface(s) 106 can include one or more ports for connecting a number of devices to one another or to another server. The memory 104 may include any computer-readable medium known in the art including, for example, volatile memory, such as static random-access memory (SRAM) and dynamic random-access memory (DRAM), and/or non-volatile memory, such as read only memory (ROM), erasable programmable ROM, flash memories, hard disks, optical disks, and magnetic tapes. In an embodiment, the memory 102 may include a database or repository. The memory 102 may comprise information pertaining to input(s)/output(s) of each step performed by the processor(s) 104 of the system 100 and methods of the present disclosure. In an embodiment, the database may be external (not shown) to the system 100 and coupled via the I/O interface 106. The memory 102, further include sentiment analysis model 110. The sentiment analysis model 110 is a reinforcement learning based model that executes language processing to assess the importance of words in the sentence for sentiment analysis. The sentiment analysis model 110 is executed by NLP module 110A and SSI module 110B. The NLP module 110 analyzes a plurality of sentiments based on information and news items available on mass communication platforms; and SSI module 110B derives a score based on the plurality of sentiments captured by the NLP module. The score is indicative of associated business reputation of the entity. The memory 102 further includes a plurality of modules (not shown here) comprises programs or coded instructions that supplement applications or functions performed by the system 100 for executing different steps involved in the process sensing alcohol intoxication of the subject. The plurality of modules, amongst other things, can include routines, programs, objects, components, and data structures, which perform particular tasks or implement particular abstract data types. The plurality of modules may also be used as, signal processor(s), node machine(s), logic circuitries, and/or any other device or component that manipulates signals based on operational instructions. Further, the plurality of modules can be used by hardware, by computer-readable instructions executed by the one or more hardware processors 104, or by a combination thereof. The plurality of modules can include various sub-modules (not shown).
[023] FIG. 2 is a functional block diagram depicting reinforcement learning involved in deriving sentiment score according to some embodiments of the present disclosure.
[024] As illustrated in FIG. 2. the system 100 receives data from the plurality of sources and processes the data through reinforcement learning wherein reinforcement learning executes NLP techniques to derive a sentiment score through keywords processed by NLP module 110B. The system 100 receives data from the plurality of data sources at 202. The data source may be unstructured or semi-structured. In an embodiment, the data source is a real time news/ wire feeds about corporate actions, regulatory actions, political actions, business, and technological actions. In another embodiment, the data source is financial data feed that includes performers, liquidity, leverage and non-performing assets (NPA) related information. In another embodiment, the data source is financial data feed comprising the information relating to price, trade volume of share, market capitalization, bonds, swaps, and the like. In another embodiment, the data source is feed on equity analyst and rating agency stance. All the data collected from the plurality of sources is pre-processed at 204, to prepare the raw data and making it suitable for the machine learning model. It is the first and crucial step while creating the machine learning model. The pre-processing involves removal of noises, missing values, and scrutinizing the unusable format which cannot be directly used for the machine learning models. Data pre-processing is a required tasks for cleaning the data and making it suitable for a machine learning model which also increases the accuracy and efficiency of the machine learning model. At 206, the pre-processed data is fed to the reinforcement learning model 110 which is a machine learning model.
[025] The reinforcement leaning is executed by extracting the keywords in the NLP module 110A. The NLP is an AI application in which the computer recognizes human language and responds to its queries. The NLP performs keyword extraction to break down human language so that it can be interpreted and evaluated by machines. It is used to extract keywords from a wide range of material, including conventional documents and business reports, social media comments, internet forums and reviews, news items, and more. During RL model training, key words are extracted from data related to the corporate actions using lists published by security dealers’ associations, stock exchanges and regulators; and the extracted keywords (such as merger, stock splits etc…) are used to train the model to pick up news on sentiments linked to such announcements. Similarly, key words were identified to train the model to pick up news on sentiments linked to management based on key words or a combo of key words such as “honored”, “awarded”, “launched initiative”, “recognized”, “person of the year”, “news maker”, “appointed chair”, “appointed member”, “government committee”, “regulatory committee”, “economic policy making body”, “industry association”, and so on. Further, key words (like “order received”, “won a contract”, “new contract” and likes) were also identified to train the model to pick up news on sentiments linked to new business initiatives or deals involving the companies under analysis. As part of reinforced learning, provisions are made to add new key words and let the model train itself based on user inputs and autonomous learning by analyzing financial, operational, market and business performance correlation with the key words introduced. The system 100 recognizes the fact that various types of news, impact the sentiment in different ways and to a different degree. The news items are categorized under various categories like financial performance, market performance, business performance, technology and innovation performance, management performance, brand performance, sustainability performance, governance risk and compliance performance. Similarly, the system 100 recognizes the fact that sentiment among various type of stakeholders has different bearing on the company. E.g., category of stakeholders includes investors, shareholders, lenders, customers, suppliers, distributors, partners, analysts’ community, regulators and policy makers, and community in vicinity of facilities and market intermediaries. Therefore, stakeholder level categorization assists the model for making more accurate predictions of sentiments. The system 100 utilizes NPL trained model for the sentiment analysis. At 208, sentiment score is derived based on news category wise sentiment score as well as stakeholder-based sentiment score relative. The weightage of news related to performance categories is assigned based on correlation analysis of past data and based on idiosyncratic assessments. However, provision is made for automatic feedback based on observed performance and sentiment score generated using the sentiment analysis model 110. User feedback provision has been incorporated to change and alter sentiment score to news category weightage relationship. Likewise, stakeholder sentiments are assigned weightages to generate a consolidated sentiment score. However, these weightages are derived based on relationship between performance linked sentiment and stakeholder sentiments. The sentiment analysis model 110 has been used to train these weightages. Provisions are also made for the user feedback to set weightages in relation to desired use of the sentiment score.
[026] FIG. 3 is an exemplary flow diagram for a method 300 for estimating financial fidelity through sentiment sensitivity index, in accordance with some embodiments of the present disclosure.
[027] The steps of the method 300 of the present disclosure will now be explained with reference to the components or blocks of the system 100 as depicted in FIG. 1 through FIG. 4. Although process steps, method steps, techniques or the like may be described in a sequential order, such processes, methods, and techniques may be configured to work in alternate orders. In other words, any sequence or order of steps that may be described does not necessarily indicate a requirement that the steps be performed in that order. The steps of processes described herein may be performed in any order practical. Further, some steps may be performed simultaneously.
At step 302 of the method 300, the one or more hardware processors 104 are configured to receive a plurality of news items from a plurality of sources wherein the plurality of sources comprises newswires, social media feeds, and news feeds on a mass-communication platforms. The system 100 captures relevant news in real-time from news feed and wire feed especially focusing on the news items related to corporate actions, regulatory actions, political context, business context or from technological context. The news item about the entity in financial context may be linked to its financial performance, liquidity, leverage capability or about the non-performing assets. The news item about the entity in a market context may be about the price, trade volume of shares, market cap or bonds swaps. The news item about the entity from equity analysts and rating agency are also considered to prepare raw data of input to the sentiment analysis model 110. The presence of the news on any mass-communication platform about an entity under analysis that can be captured from web is taken into consideration to prepare as input for sentiment analysis model 110. The system 100 performs key words search to extract all the above-mentioned news items from the web. Using the same keywords the sentiment analysis model 110 is trained that automatically fetches the relevant news items based on keywords provided. were also identified to train the model to pick up news on sentiments linked to new business initiatives or deals involving the companies under analysis. At step 304 of the method 300, the one or more hardware processors 104 are configured to categorize the plurality of news items by parsing the plurality of news items from a plurality of pre-set business rules. The pre-set business rules are made to identify all such news items which about the entity surfacing on the web with respect to one or more parameter that may be important for a lender/ financial institution to know while considering providing credits/ loan to the entity. For, e.g. the pre-set business rules may be preparing a catalogues of capturing news items based on relevant keywords. Another rule may be to focus on those news items which compute changes in values (profit, loss, sale volume etc.) and determine trend (monthly/ quarterly/ annually gain trend/ loss trend etc.). The pre-set business rules also takes care of unjustified exuberance by considering averages over a period. Sometimes, business-rules may monitor daily changes and direction of a particular news items to analyses its impact (positive or negative) on the web. Considering equity analyst feed as well as rating agency feed about the entity as the reliable and informed sentiment is also one of the business rule. At step 306 of the method 300, the one or more hardware processors 104 are configured to feed each category of the news items to a pre-trained AI/ML model. For training the AI/ML model, all the identified key words are utilized in preparing a labelled dataset. The labelled dataset is the key requirement to train the AI/ML model. The model learns various patterns in the labelled dataset and post-training phase, predicts, the sentiment for any given unseen text. Preparation of labelled dataset of the relevant key words involves text-preprocessing to cleanse the data by removing irrelevant and redundant keywords followed by numerical encoding of the text. At step 308 of the method 300, the one or more hardware processors 104 are configured to derive category-wise sentiment score of the plurality of news items. The AI/ML model classifies the news items input as labeled dataset of relevant keywords into categories. The categories are based on topics which are common among the plurality of news items and group the news items in one bucket. For e.g, news items related to financial context such as AGM report, balance sheet, quarter performance reports, SENSEX performance etc. can be grouped under one class. Similarly, news items associated from political context, e.g., geographical expansion, territorial move, government interventions (support or dissent) etc. about the entity may be grouped in another class. The basis of categorization is to recognize the fact that various types of news impact sentiment in different ways and to a different degree. The news items are categorized under various categories like financial performance, market performance, business performance, technology and innovation performance, management performance, brand performance, sustainability performance, governance risk and compliance performance. Likewise, a recognition that sentiment among various type of stakeholders has different bearing on the company, category of stakeholders like, investors, shareholders, lenders, customers, suppliers, distributors, partners, analysts’ community, regulators and policy makers, and community in vicinity of facilities and market intermediaries. The relative weightage of the news items related to performance categories are assigned based on correlation analysis of past data and based on idiosyncratic assessments. However, provisions are made for automatic feedback based on observed performance and sentiment score generated using AI/ML algorithms. User feedback provision has been incorporated to change and alter sentiment score to news category weightage relationship. Likewise, stakeholder sentiments have assigned weightages to generate a consolidated sentiment score. However, these weightages are derived based on relationship between performance linked sentiment and stakeholder sentiments. The AI/ML model has been used to train these weightages. Provision are also made for the user feedback to set weightages in relations to desired use of the sentiment score. Sentiment scores are derived for each category. The natural language processing (NLP) and artificial intelligence (AI) tools are used for sentiment score generation and further polarity determination is subjected to reinforced learning to improve the scoring. Supervised learning is used to finetune the categorization and polarity accuracy based on user’s feedback. For better accuracy in prediction, sentiment score is scaled-up to perform calculation at the higher scale. Therefore, the sentiment score is multiplied by 5 to move the scale from -1 to 1 to have it on a scale of -5 to 5. At step 310 of the method 300, the one or more hardware processors 104 are configured to specify a look-back period for the plurality of news items in each category. The look back period is specified, for which news should be looked back for determining as on date sentiment. Selection of a look-back period has a high impact on deriving more realistic sentiment score. News experience decayed sentiment based on a time scale. The recent news item have more impactful on the minds of people but as it gets older, it impact deteriorates over a period of time. Time decay factor considers look-back period and defines degrading sentiment with news farther in time back. Different time decay function such as linear, exponential etc. used for look-back period for different category, sector news item can be provided. At step 312 of the method 300, the one or more hardware processors 104 are configured to apply a time decay factor to each news item by using a decay function wherein the time decay factor represents degraded sentiments for news items farthest in time. A decay sentiment score is obtained by multiplying the category-based sentiment score of the news items with time decay factor to determine how much score to back to add on date for a news which has occurred in past. At step 314 of the method 300, the one or more hardware processors 104 are configured to adjust the category-wise sentiment score of each news item by factoring the specified look-back period and the time decay factor. Time decay factor and look-back period are relational. The farthest look-back period will have high decay factor and the nearest look-back period will have lowest decay factor. Accordingly, the news items which are quite recent will experience low decay factor wherein the older news items will experience high decay factor. The system 100 adjusts the category-wise sentiment score of each news item based on specified look-back period against each news item and the applicable time-decay factor. At step 316 of the method 300, the one or more hardware processors 104 are configured to assign a weightage to each category of the news items and multiplying the weightage with the category-wise sentiment score. To assign a weightage, category-wise sentiment score of each new item is added to derive the entity level score. The entity level score is further divided by sum of time decay factors for news item for which score is being aggregated in preceding step to normalize the consolidated sentiment score, i.e to keep the score on a scale of -5 to 5. During normalization, the news items that are not associated with any sentiment are ignored from the further calculation. In an embodiment, weightage is provisioned for various categories of the news items and hence, weightage is to be specified for each news item. However, weightage may be subject to reinforced learning. The category score for the entity is multiplied by the weightage as in preceding step and divided by the sum of weightages applied to arrive at the entity level sentiment score. At step 318 of the method 300, the one or more hardware processors 104 are configured to derive entity-level sentiment score by consolidating the category-wise sentiment score of the plurality of news items and normalize the consolidated sentiment score to sense the business reputation of the entity. The negative value of sentiment score for an entity is considered as Market Insight Pricing Index (MIPI) score for that entity. The financial fidelity of the entity is represented by sentiment sensitivity index (SSI). The SSI is dependent on many factors. One of the factors is MIPI (Market Insight Pricing Index) which has significant impact on SSI. MIPI is derived using sentiment analysis model using news feeds related to a given company. Positive and negative news to be translated in MIPI index on a scale of -5 to +5. News items (including neutral news items) are displayed as news item count by industry and news type. Except for MIPI, other SSI components can be determined using scale in lookup table and hence can be rule based. The SSI is linked to rating, risk premium, strategy pricing, relationship value, likely scale based. In an embodiment, the entity level sentiment score is further aggregated at industry level (can be further extended to other dimensions like geography etc.) based on importance of an entity to that industry sector. Importance of the entity within an industry can be based on the market share, or a more sophisticated algorithm. Therefore, sentiment scoring based SSI derived from MIPI is indicative of financial fidelity of the entity and used as an indicator by the banks/lender/ financial institution to provide credit/ financial support to the entities devoid of any credit score.

[028] SSI CALCULATION:
SSI comprises of MIPI and other factors M1, M2,……… Mn. The process of calculating SSI comprising steps,
assigning weights to each factor; and
normalizing the weighted factors by dividing weighted factor by sum of all weightages up-scaling the weighted factors to obtain SSI.
The SSI is then used to make an adjustment to the interest rate / fee fixed previously as follows:
For fixed rate,
Adjustment= Initially fixed rate *(SSI/5) * 10/100*- maximum adjustment allowed being 10 % (10/100).
[029] According to an embodiment of the present disclosure, maximum adjustment can be changed based on customers volition. The adjustment rate can be fixed. E.g. when SSI is between 0-1, the adjustment can be fixed as 2%. Similarly, when SSI is between 1-2, the adjustment can be fixed as 3%. According to an embodiment of the present disclosure, the adjustment rate can be linear or stepped linear. E.g. when SSI is between 0-1, the adjustment rate can be 0 as minimum and 2% as maximum. Similarly, when SSI is between 1-2, the adjustment rate can be 2% as minimum and 3% as maximum. According to an embodiment of the present disclosure, the adjustment can be calculated in terms of floating rate. The initial spread is adjusted appropriately. Therefore, the adjustment for floating rate is calculated as:
Adjustment= Spread Fixed Initially *(SSI/5) * 5/100 (if max adjustment is 5% or 5/100).
[030] SSI pricing has multiple other factors and MIPI is one of those factors. In an embodiment, the other factors that can be included are competitive strength wherein linear or bucketed rule-based algorithm can be used to calculate the competitive strength. Or, in case of relationship strength, multifactor algorithm can be used (e.g., value of annual/ lifetime transaction, importance of relationship in strategic terms etc.). Each of the factors in SSI model are weighted and the weighted average is then normalized to be on a scale of -5 to 5 by dividing by sum of all the weightages. Alternatively, in an embodiment, the factors in SSI model can be trained through reinforced learning.
[031] FIG. 4 is a block diagram depicting model training for performing sentiment analysis in accordance with some embodiments of the present disclosure.
[032] As illustrated in FIG. 4, the sentiment analysis model 110 undergo training for sensing relevant news and segregating the news sector-wise and category-wise. At 402, the sentiment analysis model 110 receives the plurality of news items from the plurality of sources like newswires, e-newspapers, e-magazines, news blogs, news broadcasts, social media feeds and news feeds on the mass-communication platforms. At 404, the model receives extracted keywords, topics and tags and trains to recognize keywords and tags specifying the news items and the keywords in the news items relevant for sensing business position of the entity. At 406, the sentiment analysis model also receives business information about the entity along with the plurality of news items wherein the business information comprises of financial data, market data, equity analyst data and rating agency data of the entity. At 408, the sentiment analysis model receives pre-set business rules. The pre-set business rules are based on news categories, news topic and a plurality of parameters defined for capturing sentiment score. Based on relevant keywords from news items, business information and business rules, the sentiment analysis model undergo training via reinforcement learning technique to perform sentiment analysis and thereby deriving sentiment sensitivity index.
[033] FIG. 5 is a user interface representation of the news analysis for deriving sentiment sensitivity index (SSI) in accordance with some embodiments of the present disclosure.
[034] As illustrated in FIG. 5, the plurality of news items are segregated sector-wise at 502. And for each sector, news items are categorized by setting apart each news item to a particular category. E.g., at 504, for an automobile sector, the news items are categorized under business deals, company performance, regulatory information, markets cap/debts and political categories. And sector-wise as well as category-wise count is recorded. At 506, the total number of news items considered for the SSI calculation is recorded. Further at 508, sector-wise sentiment is score is calculated by aggregating individual scores of each category. The sentiment analysis model 110 processes sector-wise and category-wise news to derives sentiment score via NLP module 110A and SSI module 110B further aggregates the derived sentiment scores to predict the sentiment sensitivity index at 510. According to an embodiment, the entity-level sentiment score is used in multitude of ways by the financial institution while analyzing financial credibility of the entity. In an embodiment, the entity-level sentiment score is used to prepare an AI/ML trained multifactor SSI for financing small and medium enterprises. The plurality of factors defining SSI include risk free rate of return, ratings, competitive ranking, aging of entity and the like. In an embodiment, rate of financing is determined based on SSI linked adjustment to base rate used for trade finance. In another embodiment, the sentiment score is utilized in depicting risk of market disruption among various stakeholders. In another embodiment, the sentiment score is utilized in estimating reputation risk to quantify transition risks.
[035] The written description describes the subject matter herein to enable any person skilled in the art to make and use the embodiments. The scope of the subject matter embodiments is defined by the claims and may include other modifications that occur to those skilled in the art. Such other modifications are intended to be within the scope of the claims if they have similar elements that do not differ from the literal language of the claims or if they include equivalent elements within substantial differences from the literal language of the claims.
[036] The embodiments of present disclosure herein addresses unresolved problem of estimating financial fidelity of the entity using sentiment sensitivity index. The credit score from reputed rating agencies are considered as a benchmark for recognition of the entity. The credit score is considered by the financial institutions evaluating the entities for lending credits/ loans. However, in the absence of credit score, there is no other parameter available to judge financial fidelity of the entity. The embodiment thus provides a method of estimating financial fidelity of the entity using sentiment sensitivity index. All the news items related to a plurality of category like financial, trade related, political, geopolitical, legal etc. appeared in the mass-communication means for a specific period of time (look-back period) are considered to derive category-wise sentiment score for the news items in each category. All such category-wise sentiment scores are assigned weightage based on look-back period and time decay factor is applied to derive the category-wise sentiment scores. The category-wise sentiment score is normalized to remove irrelevancy and redundancy and an entity level sentiment score is derived by aggregating all such normalized category-wise sentiment score. Moreover, the embodiments herein further provides SSI derived from MIPI and is indicative of financial fidelity of the entity in the absence of credit score from the rating agencies.
[037] It is to be understood that the scope of the protection is extended to such a program and in addition to a computer-readable means having a message therein; such computer-readable storage means contain program-code means for implementation of one or more steps of the method, when the program runs on a server or mobile device or any suitable programmable device. The hardware device can be any kind of device which can be programmed including e.g., any kind of computer like a server or a personal computer, or the like, or any combination thereof. The device may also include means which could be e.g., hardware means like e.g., an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or a combination of hardware and software means, e.g., an ASIC and an FPGA, or at least one microprocessor and at least one memory with software processing components located therein. Thus, the means can include both hardware means, and software means. The method embodiments described herein could be implemented in hardware and software. The device may also include software means. Alternatively, the embodiments may be implemented on different hardware devices, e.g., using a plurality of CPUs.
[038] The embodiments herein can comprise hardware and software elements. The embodiments that are implemented in software include but are not limited to, firmware, resident software, microcode, etc. The functions performed by various components described herein may be implemented in other components or combinations of other components. For the purposes of this description, a computer-usable or computer readable medium can be any apparatus that can comprise, store, communicate, propagate, or transport the program for use by or in connection with the instruction execution system, apparatus, or device.
[039] The illustrated steps are set out to explain the exemplary embodiments shown, and it should be anticipated that ongoing technological development will change the manner in which particular functions are performed. These examples are presented herein for purposes of illustration, and not limitation. Further, the boundaries of the functional building blocks have been arbitrarily defined herein for the convenience of the description. Alternative boundaries can be defined so long as the specified functions and relationships thereof are appropriately performed. Alternatives (including equivalents, extensions, variations, deviations, etc., of those described herein) will be apparent to persons skilled in the relevant art(s) based on the teachings contained herein. Such alternatives fall within the scope of the disclosed embodiments. Also, the words “comprising,” “having,” “containing,” and “including,” and other similar forms are intended to be equivalent in meaning and be open ended in that an item or items following any one of these words is not meant to be an exhaustive listing of such item or items or meant to be limited to only the listed item or items. It must also be noted that as used herein and in the appended claims, the singular forms “a,” “an,” and “the” include plural references unless the context clearly dictates otherwise.
[040] Furthermore, one or more computer-readable storage media may be utilized in implementing embodiments consistent with the present disclosure. A computer-readable storage medium refers to any type of physical memory on which information or data readable by a processor may be stored. Thus, a computer-readable storage medium may store instructions for execution by one or more processors, including instructions for causing the processor(s) to perform steps or stages consistent with the embodiments described herein. The term “computer-readable medium” should be understood to include tangible items and exclude carrier waves and transient signals, i.e., be non-transitory. Examples include random access memory (RAM), read-only memory (ROM), volatile memory, nonvolatile memory, hard drives, CD ROMs, DVDs, flash drives, disks, and any other known physical storage media.
[041] It is intended that the disclosure and examples be considered as exemplary only, with a true scope of disclosed embodiments being indicated by the following claims.
, Claims:We Claim:
1. A processor implemented method for reinforcement learning based estimation of financial fidelity of an entity, the method comprising steps:
receiving (302), via one or more hardware processors, a plurality of news items about the entity from a plurality of sources on a mass-communication platforms;
processing (304), via the one or more hardware processors, the plurality of news items by parsing the plurality of news items from a plurality of pre-set business rules to obtain category-wise news items;
feeding (306), via the one or more hardware processors, each category of the news items to a pre-trained sentiment analysis model;
deriving (308), via the one or more hardware processors, category-wise sentiment score of the plurality of news items;
specifying (310), via the one or more hardware processors, a look-back period for the plurality of news items in each category;
applying (312), via the one or more hardware processors, a time decay factor to each news item by using a decay function wherein the time decay factor represents degraded sentiments for news items farthest in time;
optimizing (314), via the one or more hardware processors, category-wise sentiment score of each news item by factoring the specified look-back period and the time decay factor;
assigning (316), via the one or more hardware processors, a weightage to each category of the news items and multiplying the weightage with the category-wise sentiment score;
deriving (318), via the one or more hardware processors, an entity-level sentiment score by consolidating the category-wise sentiment score of the plurality of news items and normalizing the consolidated category-wise sentiment score wherein the entity-level sentiment score is used in estimating the business reputation of the entity.

2. The method as claimed in claim 1, wherein the plurality of sources comprises newswires, e-newspapers, e-magazines, news blogs, news broadcasts, social media feeds and news feeds on the mass-communication platforms.

3. The method as claimed in claim 1, wherein the sentiment analysis model is trained through the reinforcement learning technique on the pre-set business rules and wherein the pre-set business rules are based on news categories, news topic and a plurality of parameters defined for capturing sentiment score.

4. The method as claimed in claim 1, wherein the look-back period is specified (i) individually to all the news items, (ii) to the category of the new items, and (iii) to a topic of the news items.

5. The method as claimed in claim 1, wherein decay function is computed using an appropriately trained function like a linear function or an exponential function for the specified look back period.

6. The method as claimed in claim 1, wherein business information about the entity is fed to the sentiment analysis model along with the plurality of news items wherein the business information comprises of financial data, market data, equity analyst data and rating agency data of the entity.

7. The method as claimed in claim 1, wherein the sentiment analysis model is trained to recognize tags specifying the news items and the keywords in the news items relevant for sensing business position of the entity.

8. A system (100), comprising:
a memory (102) storing instructions;
one or more communication interfaces (106); and
one or more hardware processors (104) coupled to the memory (102) via the one or more communication interfaces (106), wherein the one or more hardware processors (104) are configured by the instructions to:
receive, a plurality of news items about the entity from a plurality of sources on a mass-communication platforms;
process, the plurality of news items by parsing the plurality of news items from a plurality of pre-set business rules to obtain category-wise news items;
feed, each category of the news items to a pre-trained sentiment analysis model;
derive, category-wise sentiment score of the plurality of news items;
specifying (310), via the one or more hardware processors, a look-back period for the plurality of news items in each category;
apply, a time decay factor to each news item by using a decay function wherein the time decay factor represents degraded sentiments for news items farthest in time;
optimize, category-wise sentiment score of each news item by factoring the specified look-back period and the time decay factor;
assign, a weightage to each category of the news items and multiplying the weightage with the category-wise sentiment score;
derive, an entity-level sentiment score by consolidating the category-wise sentiment score of the plurality of news items and normalizing the consolidated category-wise sentiment score wherein the entity-level sentiment score is used in estimating the business reputation of the entity.

9. The system as claimed in claim 8, wherein the plurality of sources comprises newswires, e-newspapers, e-magazines, news blogs, news broadcasts, social media feeds and news feeds on the mass-communication platforms.

10. The system as claimed in claim 8, wherein the sentiment analysis model is trained through the reinforcement learning technique on the pre-set business rules and wherein the pre-set business rules are based on news categories, news topic and a plurality of parameters defined for capturing sentiment score.

11. The system as claimed in claim 8, wherein the look-back period is specified (i) individually to all the news items, (ii) to the category of the new items, and (iii) to a topic of the news items.

12. The system as claimed in claim 8, wherein decay function is computed using an appropriately trained function like a linear function or an exponential function for the specified look back period.

13. The system as claimed in claim 8, wherein business information about the entity is fed to the sentiment analysis model along with the plurality of news items wherein the business information comprises of financial data, market data, equity analyst data and rating agency data of the entity.

14. The system as claimed in claim 8, wherein the AI/ML model is trained to recognize tags specifying the news items and the keywords in the news items relevant for sensing business position of the entity.

Dated this 7th Day of December 2023

Tata Consultancy Services Limited
By their Agent & Attorney

(Adheesh Nargolkar)
of Khaitan & Co
Reg No IN-PA-1086