FIELD OF INVENTION;

The present invention generally relates to computer implemented methods and systems for enabling decision making based on automatically generated insights from analysis of large amounts of unstructured user feedback about any entity(e.g. Product, person, organization or place) and its sub-entities (e.g. Features, related entities) collected from online and offline sources. More particularly, the present invention relates to a method and system which provides insights automatically for different decision contexts by processing unstructured information from both online and offline sources.

BACKGROUND OF THE INVENTION;

People need to make several decisions for their daily needs. Consumers need to decide which products best satisfy their needs and desires. Travellers need to decide which place to visit or which hotels to stay at. Voters need to decide which politician to vote for. Organizations need to understand which of their products or features are liked by users for product planning. Traditionally these decisions were made based on structured information like price, attributes, specifications or rating systems, but a wealth of information is hidden in user generated content on social media sites, blogs, websites and news articles. Given the large number of places where people state their feedback and the fact that this feedback is stated in natural language which machines cannot understand, it has been impossible to get a consolidated view of user feedback for any entity and its sub-entities. Going through the feedback manually to understand entity and sub-entity level sentiment, co-relating between various concepts being discussed and deriving a statistical number based on which decision can be made is time consuming and expensive and could be biased to the person reading and summarizing. It also requires the person to have a very good understanding of the domain .

In general, there are several methods or systems available for providing insights for decision making. Most of the conventional systems rely on five star rating that may be misleading. Collection and ranking of the products based on sentimental analysis are well known in the prior art. For example, U.S. Pat. No. 8,117,207 to Mushtaq et al., describes a method for evaluating feature opinions for products, services and entities. This method analyses reviews, generates opinion score from positive, negative and neutral reviews utilizing information extractor and sentiment rating engine. A non-sense analyzer rejects the repeated opinion. Also, the method aggregates sentiments (product/feature) for generating scores. U.S. Patent Application No. 20090319342 to Shilman et al., describes a system and method for aggregating and summarizing product sentiment. This method involves the collection of product information, product specification, price information, blog post and forum post from reviews. The collected information is processed for ranking several products using sentiment and relevance analyser. However, this prior art does not mention sentiment analysis at sub-entitiy (e.g. Feature) level which is very important for decision making as well as other essential activities like removing duplication, spam filtering etc. Wwhich improves the quality of content.

U.S. Pat. No. 7,974,983 to Goeldi et al., discloses a method for assigning sentiment rate/score to each of the keywords related to products, services and features using sentiment rating processing module. This method includes a training or feedback loop where the keywords may be re-rated over time based on experience. However, the above mentioned prior art does not disclose a fully automated system and repeatable technique for generating insights from large volumes of unstructured user feedback for decision making which can be applied to any domain using unique techniques like ad/spam and duplication filtering, advanced natural language processing techniques like ontological anaysis, use of declensions, grammer pattern based rules, n-grams based concept mining and Subject-Verb-Object analysis, domain specific decision framework, automatic generation of ontologies, represnetaion of sentiment as positivity/negativity, buzz and personalization concepts like user-opinion holder matching, opinion scoring and ranking techniques, personal shelf, filtering based on trust etc. Hence, there is need for a new and fully automated system for generating insights for different decision context from unstructured user feedback from offline and online sources. This will enable the users to get a comprehensive, unbiased and personalized understanding of wisdom of the crowds before taking decisions.

SUMMARY OF THE INVENTION:

An object of the present invention is to provide be-spoke insight framework for aiding customer to make decisions of every context, based on automatically generated insights from unstructured user feedback collected from all sources, both online and offline. The process of the present invention includes four steps: 1) Reading 2) Understanding and 3) Ranking 4) Personalized delivery According to the present invention, the first reading step includes extracting or harvesting the unstructured data (Text/Videos/Audio etc.) using crawlers or connectors to sift through social media websites, review sites and other web pages to find discussions related to entities of interest. Cleansing engine of the present invention weeds out the repeated comments, ads and offers, smape etc. to prepare good quality content for further processing.

The second understanding step uses a combination of NLP (Natural Language Processing) linguistic rules, Statistical Analysis, Machine learning, Semantic concepts (e.g. Ontologies) and Big-data technologies (e.g. Map-Reduce, NoSQL) to store and process large volumes of unstructured content like text, images, videos etc. These steps extracts excerpts of interest, classifies them and tags them for e.g. Sentiment and type, comparision, wants ans wishes etc. The final ranking step stores the information in a graph data structure to establish relationships between entities and then applies statistical methods to generate various insights for decision making like positivity / negativity score and buzz score, competition etc.An interactive user interface of the present invention presents the insight to the user and offeres various personalization capabilities like sorting, ranking based on user-opinion holder matching, user preference matching etc.

BRIEF DESCRIPTION OF THE DRAWINGS:

The objective of the present invention will now be described in more detail with reference to the accompanying drawing.

Fig. 1 illustrates the process involved in generating insights for decision making from unstructured user feedback collected from online and offline sources.

DETAILED DESCRIPTION OF THE INVENTION:

The present invention relates to a system and method that provides automatic generation of insights from various decision contexts from unstructured user feedback collected from online and offline sources. Referring to the invention in detail, fig.l illustrates the steps involved in decision making framework according to the present invention. The system includes data collection module 1, distillation module 2, ontology 3, opinion extraction module 4, graph storage module 5, ranking and scoring module 6, decision making framework 7, insight generation module 8 ,user interface 9 and personalization engine 10. Data collection module 1 extracts unstructured data (Text / Videos / Audio etc.) from various social media websites, review sites, web pages etc. and offline sources using connectors or A crawlers, file loaders and convert into a standard format for processing. The standard format includes extracted content and metadata like author, date, source and url. The extraction process is extensible to allow new sources to be plugged-in in the future. Distillation module 2 filters the content collected from various sources both online and offline to prepare good quality content for further processing. Further, the distillation module 2 filters the collected content using various forms of filters. The various filtering forms include spam filter, ads and offers filter, de-duplication filter, removal of unrelated sentences or relevance filter and profanity filter.

Spam filter filters spam content using text processing to identify promoted content based on words used and frequency of appearance. Ads and offers filter removes ads and offers using text processing rules to identify words and phrases. De-duplication filter generates min-hash for each piece of content and compare the hash to identify similar content. Min-hash technique enables this to be done on a parallel computing platform. Relevance filter uses ontological analysis to remove content which does not contain any information of interest. Profanity filter uses gazetteer rules to identify content which contains profanity and removes it. Domain of interest is modeled to create ontology 3 of entities, sub entities and relationships which exist in that domain along with synonyms. E.g. Products and its features, Party and its politicians, place and its hotels, Hotel and its features etc.. This information is used by text processing to understand the text better. Information extraction module 4 extracts opinions on a parallel computing platform using the following techniques:

• Tokenization: It is a process in which text is split into words, phrases, symbols, or other meaningful elements called tokens. The tokens are used as an input for further processing

• Word correction (Regex): In this process the mis-spelt and slang words are corrected into valid English words using regular expressions

• Sentence splitting: It assembles the tokenized text into sentences and paragraphs using various techniques like line breaks, sentence patterns, punctuations etc. It also split compound sentences (including those with referential pronouns) based on connectives (along with the exception list)

• Mark paragraphs: Paragraphs are marked based on line breaks

• POS tagging: Part of speech tagging is the process to identify adjectives, nouns, verbs etc. using machine learning as well as linguistic rules. Two-phase POS-Tagging is used to address the issue of nouns in verb forms and other such irregular forms

• Morphological analyzer: It is a program for identifying root of every word using stemming and dictionary lookup techniques

• Named-entity recognition (NER): It is the process locating named entities and phrases using gazetteer look-ups and noun-verb analysis

• Ontological analysis: It uses domain ontology to mark entities and sub entities

• Combined Nouns: Combined nouns are marked, i.e. nouns which make sense together

• Domain relevance check: It is the process to check if the content is of relevance to the domain of interest using ontological analysis and statistical analysis

• Pronominal Co-referencer: It identifies the nouns to which the pronouns refer to using techniques like proximity, number and persona agreement

• Mark questions: Questions are marked based on grammatical patterns and punctuations

• Wants and Wishes: Mark sentences where somebody has expressed a want or wish from some other entity (target). Techniques like Named Entity Recognition and Subject-Verb-Object analysis are used for this purpose.

• Mark quoted text: Quoted text is marked based on grammatical patterns and punctuations

• Mark Declensions: Case ending or declensions are marked in which different nouns are used

• Mark Clauses: Dependent or Independent clauses are marked

• Mark Phrases: Common phrases are marked using gazetteer look-ups

• Key-phrase extraction: It is the process which extracts key-phrases through statistical analysis

• Mark Senti-words: Words which signify sentiment are marked using gazetteer look-ups and modifiers like anit-words, phrases, action, sentence types etc

• Mark subjects: Subject of a sentence is marked using declensions, clauses, grammatical patterns, statistical analysis and pronominal co-reference

• Sentence types: Different sentence types are marked as Conditional/Predictive/Assertive using grammatical patterns and gazetteer look-ups

• Opinion Holder Extraction from text: This method identifies the entity, who holds the opinion. Techniques used are Reported Speech analysis and Named Entity Recognition

• Concept mining : Mine concepts i.e. topics from User comments using methods such as Key-Word lookup, N-Grams, and Latent Semantic Indexing.

• Mark Potential Heads: Potential heads of the opinion are marked using techniques like entity recognition, references and subject-object analysis

• Review info extraction: It is the process which analyzes pros/cons section in review documents

• Mark Opinions: Opinion snippets are marked using techniques like grammatical patterns, Subject /Object /Verb analysis and positional patterns

• Mark Comparisons : Mark entities being compared.

• Resolve Head of Opinion: Head of the opinion is resolved using gazetteer look-ups (adjective-noun linkages), Units based, Frequency based and Paragraph Heading based rules

• Opinion Linking to Entities: opinions are linked to entities using techniques like proximity based, sentence/paragraph boundary based, Statistical analysis based topic identification, original search term, ontological analysis (e.g. A car has tires and steering wheel while a cell phone has screen and battery) and linguistic analysis (usage of words).

• Wants and Wishes Linking to Entities using techniques like proximity based, sentence/paragraph boundary based, Statistical analysis based topic identification, original search term, ontological analysis (e.g. A car has tires and steering wheel while a cell phone has screen and battery) and linguistic analysis (usage of words).

• Questions linking to Entities using techniques like proximity based, sentence/paragraph boundary based, Statistical analysis based topic identification, original search term, ontological analysis (e.g. A car has tires and steering wheel while a cell phone has screen and battery) and linguistic analysis (usage of words).

• Comparisons linking to entities using techniques like proximity based, sentence/paragraph boundary based, Statistical analysis based topic identification, original search term, ontological analysis (e.g. A car has tires and steering wheel while a cell phone has screen and battery) and linguistic analysis (usage of words).

Graph storage module 5 stores extracted content into a graph establishing relationships between entities, sub entities, opinions, locations, and users. Storing data in graph instead of relation storage enables adding new domains in the future and in making the queries flexible. It also enables reasoning to identify new entities and relationships automatically. Ranking and Scoring module 6 rank and score entities/topics and sub entities/topics based on positivity/negativity and buzz using statistical methods (Modified Bayesian averages). Buzz score is calculated based on how widely people are talking about a product or topic and how much an opinion can be trusted. The Modified Bayesian averages factoring in parameters like number of opinions, emotion expressed in the opinion and strength of emotion, age of the opinion, relative comparison to similar entities, geographical spread, bayesian averages and weightages are calculated.

Decision making framework 7 generates insight categories and sub topics required for decision making for a given domain using manual and automated processes. Framework includes defining parameters that help in decision making, define filter criteria for entities, generating comparison parameters and creating persona driven variations for decision making framework. Insight generation module 8 identifies insights required for decision-making within each insight category and generates insights from ranks, scores and facts of entities/topics. A rich interactive user interface 9 present the insights which makes it easy for user to see the information and make decisions. The UI mainly consists of three sections: (1) Discovery, where the user can discover and narrow down on an option based on various parameters like filters, needs, personas etc (2) Entity hero page, where all information related to the entity is shown in one place to help the user make a decision and (3) Comparison, where user can compare different entities on various parameters.

A personalization engine 10 which personalizes the information shown to the user based on various parameters like:

• User - Opinion holder matching based on parameters like demographics or explicit user preference

• Ability for user to filter/sort based on various parameters like relevance, date, source, type of media (text, video, image) etc.

• The weightages can be specified by the user for sources e.g. social/expert, opinion holder (e.g. Friends/experts), sub entities/sub topics of interest (e.g. camera, direction etc.), persona (e.g. Traveller, businessman) and user filtered entities/topics of interest (e.g. 3G, actor Tom Hanks etc).

• Ability for users to store entities of interest to a "Shelf to quickly access them and to get updates related to them.

• Ability for users to connect to social networking sites to ask feedback from friends or crawl their private network to get insights.

While the foregoing written description ofth e invention enables one of ordinary skill to make and use what is considered presently to be the best mode thereof, those of ordinary skill will understand and appreciate the existence of variations, combinations, and equivalents of the specific embodiment, method, and examples herein. The invention should therefore not be limited by the above described embodiment, method, and examples, but by all embodiments and methods within the scope and spirit of the invention as claimed.

WE CLAIM:

1. A system and method to provide an decision making framework for differen contexts based on insights generated from large amounts of unstructured user feedback from online and offline sources comprising of: data collection module 1; distillation module 2; ontology 3; information extraction module 4; graph storage module 5; ranking and scoring module 6; decision making framework 7; insight generation module 8; user interface 9 and personalization engine 10.

2. The system according to claim 1, wherein said data collection module 1 extracts the unstructured data using crawlers or connectors and converts into a standard format.

3. The system according to claim 1, wherein said distillation module filters 2 the collected content using various forms of filters.

4. The system according to claim 3, wherein said distillation module filters 2 include a spam filter, ads and offers filter, de-duplication filter, relevance filter and profanity filter.

5. The system according to claim 1, wherein said ontology 3 for a domain of interest is used to mark entities, sub entities and relationships.

6. The system according to claim 1, wherein said information extraction module 4 extracts opinions on a parallel computing platform using (i) Tokenization; (ii) Word correction (Regex); (iii) Sentence splitting; (iv) Mark paragraphs; (v) POS tagging; (vi) Morphological analyzer; (vii) Named-entity recognition (NER); (viii) Ontological analysis; (ix) Combined Nouns; (x) Domain relevance check; (xi) Pronominal Co-referencer; (xii) Mark questions; (xiii) Mark wants and wishes; (xiv) Mark quoted text; (xv) Mark Declensions; (xvi) Mark Clauses; (xvii) Mark Phrases; (xviii) Key-phrase extraction; (xix) Sentence types; (xx) Mark Senti-words; (xxi) Mark subjects; (xxii) Mark Sentence types (xxiii) Opinion Holder extraction (xxiv) Concept Mining (xxv) Mark Potential Heads; (xxvi) Review info extraction; (xxvii) Mark Opinions; (xxviii) Resolve Head of Opinion; and (xxix) Opinion Linking to Entities, (xxx) Link wants and wishes to entities (xxxi) Link questions to entities (xxx) Link comparisons to entities.

7. The system according to claim 1, wherein said graph storage module 5 stores the opinions into a graph establishing relationships between entities, sub entities, opinions, locations and users as well as identifies new entity types and relationships through reasoning.

8. The system according to claim 1, wherein said ranking and scoring modules 6 rank and score entities/topics and sub entities/topics based on positivity/negativity and buzz score using statistical methods.

9. The system according to claim 8, wherein said buzz score is calculated based on how widely people are talking about an entity and how much an opinion can be trusted.

10. The system according to claim 8, wherein said statistical methods include Modified Bayesian averages.

11. The system according to claim 10, wherein said Modified Bayesian averages factoring in parameters including number of opinions, emotion expressed in the opinion and strength of emotion, age of the opinion, relative comparison to similar entities, geographical spread, Bayesian averages and weightages, popularity of the opinion holder and popularity of the opinion itself (likes, shares).

12. The system according to claim 11, wherein said weightages can be specified by the user for sources, opinion holder, sub entities/sub topics of interest, persona and user filtered entities/topics of interest.

13. The system according to claim 1, wherein said decision making framework 7 generates insight categories and sub topics required for decision making for a given domain using manual and automated processes.

14. The system according to claim 13, wherein said decision making framework 7 further includes defining specifications that help in decision making, defining filter criteria to entities, generating comparison parameters and creating persona driven variations.

15. The system according to claim 1, wherein said insight generation module 8 identifies insights required for decision-making within each insight category and generates insights based on ranks, scores and facts of entities/topics.

16. The system according to claim 1, wherein said user interface 9 presents the insights which make it easier for user to consume the information and make decisions.

17. The system according to claim 16, wherein said user interface 9 includes discovery, entity hero page and comparison sections.

18. The system according to claim 17, wherein said discovery section is used to discover and narrow down on an option based on various parameters like filters, needs, personas by the user.

19. The system according to claim 17, wherein said entity hero page section includes all information relating to the entity in one place to help the user in making a decision.

20. The system according to claim 17, wherein said comparison section is used to compare different entities on various parameters by the user.

21. The system according to claim 1, wherein said personalization engine 10 personalizes the insights shown to the user based on various parameters like user-opinion holder matching, sorting, filtering, storing in shelf and connecting to users private social networks.

22. A method for providing an decision framework for different contexts based on insights derived from large amount of unstructured user feedback from online and offline sources, the method comprising the steps of: collecting contents from publically available resources; distilling the said collected content using various forms of filters; modelling the domain of interest to create an ontology of entities, sub entities and relationships which exist in the said domain; based on the said ontology, segmenting the contents based on the said domain of interest; extracting opinions on a parallel computing platform from the said segmented content; storing the said extracted opinions in a graph to establish relationships between said entities; applying statistical methods to said extracted opinion to generate positivity / negativity score and buzz score to rank the products based on the said ranks and scores; generating insights required for decision making; and presenting the said insights about the products to the consumer in an interactive user interface and personalizing the insights to user profile and needs.