FIELD OF DISLCOSURE
The present disclosure refers to a system and a method for product feature discovery and ranking using a 3 stage approach.
BACKGROUND
The advent of social media has led to exponential amounts of unstructured data piling in the recent past.
Typically, a product is tracked based on the parameters defined by its vendor. It has become very important for various organizations and companies to track the performance of their respective brands vis-a-vis their rivals, the rate of growth only compounding the hurdles in achieving the required. Also due to humongous volumes of data, companies are uncertain about the parameters they should focus on, which invariably leads to listing of and evaluating a wrong parameter or at most times, missing out a critical parameter.
These parameters are defined as a part of specification i.e. words or keywords that define/describe a product or an object, often being expressed in several ways. The challenging aspect of social media is that the keywords entered to search parameters of a certain product may not match the opinions expressed as defined by the vendor. Also, the opinions could be expressed on the features which are not yet identified by the vendors. All this poses a tremendous challenge in identification and matching of ail the required and germane features.
To manually perform an evaluation of reviews and opinions and then classify the parameters based on human understanding may be an ideal approach, but it, needless to mention would consume considerable amount of time, money and

effort. Another approach could be to go by the list of defined parameters and automate the extraction. This however will not yield accurate results as a lot of content will go unnoticed because expressions are not limited to the mention of such parameters.
It has therefore becomes a daunting problem for companies to keep pace with and keep a tab on such a disproportionate and unstructured growth. A need henceforth is felt, to identify and compare all tracking features vis-a-vis competitors. This can help the companies to focus on the current business critical features which are mentioned in different online sources. Owing to a disjointed and incoherent nature of the content, the identification and the description of the relating features mentioned in different sources become non-uniform. This fails to render a comprehensive set of common features to compare products and objects. A considerable amount of research needs to be done to identify commonality to find the best out of the lot. For e.g.: Tool evaluation to identify a product suiting the requirement.
We also need an internal enterprise system to mine the features. This can help in improving research and development and also for charting future business strategies associated with the product.
SUMMARY
The present disclosure teaches various embodiments of the invention, wherein an approach for building a system is presented that automatically extracts the critical set of product/object features and parameters from a wide compass of facts and opinions.

Although many approaches have been developed for extracting features from text, our focus here is to extract the critical set of features, by mapping those extracted from facts with those extracted from opinions. Making use of online sites with author-generated comments or opinions, the system aligns the features from factual online sources with those found from social media. A validation and mapping model is then proposed over the resulting data set, to subsequently rank the important features. This concept helps companies and ■■ organizations to focus on only those that matter the most for the business while also monitoring brands in Social media.
Facts, in this specification, mean objective expressions about objects, brands, products and other properties. Opinions, in the specification, mean subjective expressions that describe people's sentiments, appraisals or feelings towards those objects, brands, products, people, events and their other properties.
In one embodiment of the invention, it extracts elements from factual database and opinion database which may be factual online sites like Wikipedia, Factiva and the likes; and opinion online sites like product review sites, blogs, forums and the likes. Since the social media is involved, the system is capable of extracting features that are more popular among the masses. In achieving this it uses the combination of 'double propagation', 'part-whole' and 'no' patterns inspired and culled out from the existing state of art.
Another aspect of the disclosure teaches that the parameters or the features of the facts and opinions are two in it kinds, namely implicit and explicit. The explicit features are the one that are conspicuous and evident in a sentence, the meaning of which does not have to be construed. The implicit features are the one that are subtle and not apparent in a sentence, the meaning of which has to be construed after some comprehension and deduction.

An illustration for the above is provided here. For instance if we say, 'the ruler is long'; the implicit feature is the 'size'. Though the word 'size' does not feature in the sentence but it can be understood from the sentence that an attribute/feature ' size' is being talked about. In another instance if we say, 'the touch screen is good'; the explicit feature here is the touch screen because it figures in the sentence and is also the attribute/feature being talked about.
In the same embodiment, the feature extraction makes use of crawler and schedulers. A crawler is a computer program that browses the World Wide Web in a methodical, automated manner or in an orderly fashion. A scheduler takes care of the date set extraction and is triggered at regular intervals. Thus the feature set extracted can be made available at more real time blasts.
In another embodiment of the invention the implicit features are identified using a 'feature identifier'. The feature identifier makes use of an interpretation dictionary consisting of different adjectives, adverbs, adjective adverb combinations. Implicit features are then extracted and finally put into a factual data set.
In the same embodiment the parameters/features mentioned directly in database are opinions and aggregate the 'explicit features' into an opinion data set. The feature identifier extracts implicit features and adds it to the opinion data set.
In another embodiment of the invention, the extracted features from the factual database and opinion database are mapped using the 'validation and mapping model'. This is done by building a mapping configuration between different features of any given product and objects by extracting data from factual online

sources such as encyclopedias, product specifications, catalogs etc with those extracted from social media, it extends the concept further by providing a validation and mapping mechanism for the features that are important for the product/object.
Yet another embodiment of the invention teaches a mapping mechanism which helps in identifying features that don't exist in facts but are mentioned in opinions and vice versa, thus helping in identifying the maximum possible number of features and parameters possible.
In the same embodiment, the mapping mechanism is done not just by 'word comparisons', but by 'semantic relations' using 'semantic intelligence'. The mapping rules either cater to aggregating words such as unigrams, bigrams or trigram based on patterns and/or rules containing static texts which represent product/brand features and/or rules that may look for group of words containing specific semantic patterns for e.g.: map only if the word/words occur in a sentence containing a specific noun/adjective before it and/or map if the word finds mention along with one of the brand or product of its competitor and/or for the synonyms of words and/or for the partial matches of synonyms or Hyponyms.
In a different embodiment of the invention the factual and opinion data sets contain implicit and explicit features from the respective data sources. The mapping mechanism gives validity to each of the extracted features and the product and objects have a dynamic set of features based on current trends on social media with improved accuracy in features as the data is validated with facts.
In the same embodiment of the invention the mapping performed is of two kinds, namely implicit and explicit. Based on the feature-indicator and synonym

configuration, the system of the instant invention will perform a mapping between factual features with one or more opinion features and vice versa. The implicit mapper will use an "implicit mapping" configuration which has all possible synonyms that refer to a feature. This configuration will continue to grow based on newer feature-synonyms availability. On the other hand, all those features which cannot be mapped by the system, is presented to the user in a User Interface. They are words that did not find mention in the implicit mapping configuration used by the "Implicit mapper" mentioned above. Users can either add these words and map it to existing ones in the configuration or perform an explicit mapping between the two sets. This will then be used by the system for future feature set identification and will be used to train the system for higher precision.
According to another embodiment of the disclosure, the features mapped are ranked in order of their importance. This is done using feature relevance and feature frequency. The emphasis on the approach here is to ensure that the features are more relevant to the feature discovery for the customers after extraction. It uses an already existing feature extraction algorithm to extract the features from the factual and opinion database. The features that have a higher rank appear on the top of the list.
The various embodiments of the invention teach to derive a method and system for discovery and ranking the list of brand features that make a difference to the product at any given point of time. It finds application in domain specific dictionaries for sentiment analytics systems, in social publishing where the vendors can identify consumers based on their comments on specific features and can publish relevant content accordingly. It has its usage in feature based brand tracking and in social marketing to understand users commenting on new features.

BRIEF DESCRIPTION OF DRAWINGS:
The detailed description is described with reference to the accompanying figures. In the figures, the left-most digit(s) of a reference number identifies the figure in which the reference number first appears. The same numbers are used throughout the drawings to reference like features and components.
Figure 1 illustrates a block diagram representation for a system for
discovery using a 3 stage flow.
Figure 2 illustrates a flow diagram representation for a concept for
discovery using a 3 stage concept flow
Figure 2(a) illustrates a flow diagram for feature extraction.
Figure 3 illustrates a flow diagram representation for implicit in
accordance with an embodiment of the present disclosure.
Figure 4 illustrates a flow diagram representation for explicit mapping
in accordance with an embodiment of the present disclosure.
Figure 5 illustrates a mapping representation between the factual data
set and the opinion data set.

DETAILED DESCRIPTION:
The following discussion provides a brief, general description of a suitable environment in which various embodiments of the present disclosure can be implemented. The aspects and embodiments are described in the general context of computer executable mechanisms such as routines executed by a general purpose computer e.g. a server or personal computer. The embodiments described herein can be practiced with other system configurations, including internet appliances, hand held devices, multi-processor systems, microprocessor based or programmable consumer electronics, network PCs, mini computers, mainframe computers and the like. The embodiments can be embodied in a special purpose computer or data processor that is specifically programmed configured or constructed to perform one or more of the computer executable mechanisms explained in detail below.
Figure 1 illustrates a system for automatic feature discovery and ranking of a product using a 3 stage approach in the instant invention with factual data module (101), execution module (102) and opinion data module (103).
The factual data module (101) is configured to prune down features and parameters from different factual data sources and factual databases containing all possible facts of all possible parameters and features for all possible products or objects. The factual data sources or factual databases may be online sites like Wikipedia, Factiva or encyclopedias which may be online or offline. Controlling access to the factual database and arranging the extracted features in order of significance for further extraction is also a function entrusted with the factual data module (101).

The opinion data module (102) is configured to prune down features and parameters from different opinion data sources and opinion databases comprising all possible feedback or appraisal for all possible products or objects, the feedback or appraisal either made available on online review sites, blogs or made available on forums or an opinion database. Controlling access to the opinion database is also the prerogative of the opinion data module (102).
The data from the factual module (101) and the opinion module (102) is then fed to the execution module (102) for further processing to take place. In one embodiment of the invention the execution module (102) comprises a factual data adapter (104). The data from the factual module (101) is fed to the factual data adapter (104) configured to extract the relevant and the most prudent features or parameters of particular product or an object under scanner. The factual data adapter (104) approaches the situation with 'double propagation', 'part-whole' and 'no' pattern relations using an already existing feature extraction algorithm.
In another embodiment of the invention the opinion data adapter (105) ,residing inside the execution module(102) /take its inputs from the opinion data module (103) to process and extract the features or parameters, whichever it finds the most germane for the particular product or the object under scanner. To aid and facilitate the process of extraction it also makes use of feature extraction algorithm and the approaches of 'double propagation', 'part-whole' and 'no' pattern relations.
in another aspect of the invention the execution module (102) contains a data assembler module (106), which assembles the content it receives from the factual data adapter (104) and from the opinion data adapter (105). It arranges the

content conflated from both 104 and 105 in a priority based orderly manner with due prominence on their relevance and importance.
A feature disambiguator, in another embodiment is provided that resides inside the execution module (102), takes the data from the data assembler (106) and identifies explicit and implicit features within the test using a feature extraction algorithm, already known to the existing state of art.
In another aspect of the invention the execution module (102) also contains a feature mapping engine (108) tracks, observes and identifies patterns and maps features between the data collocated from the feature disambiguator (107) and provide its analysis back to the disambiguator (107).
Within the feature mapping engine (108) an implicit and explicit mapping is performed which can be one to one, one to many, many to one and many to many. Whereas the explicit mapping involves a new round of manual mapping of unmapped entries and uninitiated features at regular intervals, the implicit mapping executes in a self improvisation automated fashion employing the use of an 'implicit mapper' to suit its purpose. At this instant, the invention talks about self learning, not just by 'word comparisons' but by using semantic relations, embarking on the concept of semantic intelligence. More accentuation on such derived mapping rules are dealt later in the specification.
Herein one or more factual features of one set of factual data adapter (104) are compared with one or more features taken from the opinion data adapter (105) using an 'implicit mapper' .The implicit mapper is a technique that creates a dynamic set of features which continues to grow and refine itself contingent on the newer outcomes, the consequent domino effect and the extent of usage of the feature identifier (109).

The execution module (102) also contains a feature identifier (109) which identifies and compares features between the factual content and opinion content obtained from the feature disambiguator (107).
While performing this, the implicit feature identifier (109) makes use of an interpretation dictionary, which consists of different parts of speech containing synonyms to the compared feature between factual data adapter (104) and opinion data adapter (105). Due relevance to each pattern in the dictionary is allocated, with simultaneously comparing the older to newer features extracted from same data source and determining whether the feature were extracted earlier or not.
The feature disambiguator (107), the feature mapping engine 91080 and the feature identifier (109) control the inflow and outflow of their input and their output regularly to finally pass on the result to a data store (110).
The data store (110), in yet another embodiment of the invention resides inside the execution module (102) and stores features or parameters for each product or object under study.
The features and the parameters thus obtained are ranked based on feature relevance and feature frequency.
The execution module (102) containing a publisher (111) at various instant of request by the end user is configured to possess the intelligence to comprehend features or parameters and publish into the social media based on business rules. The output from the data store (110) is fed to the application and services (112) finally.

Figure 2 shows a schematic representation of an exemplary method (200) for performing a 3 staged product feature discovery and ranking, the implementation of which, at this instant, is done by the factual data adapter (104) ,the opinion data adapter (105) and by the data assembler (106) ,all residing inside the unit of execution module (102).
The method is illustrated as a logical flow graph, which represents a sequence of operations that can be implemented in hardware, software, or a combination thereof.
In one embodiment of the method a list of product or object relevant factual features is identified from factual data sources or factual database (201) The factual data sources or factual database have been elucidated in much detail in 101.
These relevant features are further pruned down and the search is refined (202). The factual features thus extracted are queued up and stored in a factual data set (203).
In another embodiment of the method a list of opinion features is identified from opinion data sources and opinion database (204).
The opinion data sources and opinion database have been elucidated upon in much detail in 103.
These relevant features are further pruned down and the search is refined (205). The opinion features thus extracted are queued up and stored in a factual data set (206).

Another aspect of the method teaches validation and mapping of the features obtained from the 1st and 2nd embodiments of the method mentioned above. The mapping can be one to one, one to many, many to one and many to many. Implicit mapping is performed (207) inside a feature mapping engine unit (108) with an implicit mapper wherein the features are mapped using an interpretation dictionary, and explicit mapping is performed (208) by manual intervention, the resultant of which is stored in the final list of features (209). The two mappings interchangeably use the feature identifier unit.
In yet another embodiment of the method the features are subsequently ranked based on defined ranking rules, feature relevance and feature frequency. (210).
Referring to Figure 2{a), shown is a flowchart of an example method of extracting factual features (202) and extraction of opinion features (205) from the factual and opinion databases respectively by mean of the factual data adapter unit (104) and also by opinion data adapter unit (105). At 211, it is determined whether the list coming from 202 is null /empty. If the list indicates otherwise, a next URL is queued up at 212, and the corresponding document is searched for at 213. This searched document is split into sentences at 214. The sentences are identified and subsequently queued up( 215 and 216). The techniques of 'double propagation', 'part whole' and 'no 'patterns, already known to the existing state of art, are applied at 217. Post this, the result is added to the extraction list at 218. Features indicator(109) refers to the interpretation dictionary at 219 residing inside the feature mapping engine unit (108), the same is then added and stored in the feature set (220 and 221) respectively by means of data assembler unit (106)

The feature indicator employs the use of interpretation dictionary (222) availed inside the feature mapping engine unit (108) which contains all possible word combinations, adverbs, adjectives and adverb -adjective combinations to match with the word confronted. After the feature extraction at every stage the control is given back to the next stage. The result is added to and stored In the extraction list by means of data store unit (110)
Figure 3 illustrates an embodiment of the instant invention regarding an implicit mapping approach by means of the execution module unit (102), more precisely and significantly the feature mapping engine unit (108) and the feature identifier unit (109).
The implicit mapping configuration is loaded (301), post which the implicit mapper (302) is loaded too, both by means of feature mapping engine unit (108). The implicit mapper (302) receives its inputs from the factual data set (309) by means of the factual data adapter (104) and the opinion data set (308) by means of the opinion data adapter (105).
Thereafter with the help of feature identifier unit (109) a list A is obtained from the factual data set (309) and is compared to a list B obtained from the opinion data set (308). This comparison entails the use of an interpretation dictionary, which consists of different parts of speech containing synonyms to the compared feature between factual data adapter (104) and opinion data adapter (105). Due relevance to each pattern in the dictionary is allocated, while simultaneously comparing the older as newer features extracted from same data source and determining whether the features were extracted earlier or not.

The mapping mechanism model exploits semantic intelligence and semantic relations and is not restricted to word comparisons. The mapping rules either cater to aggregating words such as unigrams, Digrams or trigram based on patterns and/or rules containing static texts which represent product/brand features and/or rules that may look for group of words contain specific semantic patterns for e.g.: map on(y if the word/words occur in a sentence containing a specific noun/adjective before it and/or map if the word finds mention along with one of the brand or product of its competitor and/or for the synonyms of words and/or for the partial matches of synonyms or Hypernyms.
In an indication, with due cognizance gained from the interpretation dictionary, that list A exists; the items in the list are parsed and mapped (304 and 305) onto the items in the list 8 by means of a feature mapping engine unit (108). On determining whether the match is available for A or not (306), and if the availability succeeds, then the matched features are collocated (307) and implicit stored (308) for the list A are stored by means of data store unit (110).
Figure 4 illustrates a flow diagram for explicit mapping executed by means of the execution module (102), primarily by the feature mapping engine (108). At 401, the factual data set (203) and the opinion data set (206) are loaded onto the feature mapping engine (108) by means of a feature disambiguator (107). The loading of implicitly mapped features are however excluded from the feature disambiguator at this stage. The data sets are displayed to the user so that the unmapped entries and features can be identified (402) and manually added {403 and 404) to the list. The status of a requirement to perform implicit mapping is determined (406) by a particular field, which is set to 'No' by default and gets changed to 'Yes' on loading the implicit mapping configuration (301).

Figure 5 illustrates (500), in the same embodiment as above, a mapping between the factual data set (203) and the opinion data set (206).
The present invention is not to be limited in scope by the specific embodiments and examples which are intended as illustrations of a number of aspects of the invention and all embodiments which are functionally equivalent are within the scope of this invention. Those skilled in the art will know, or will be able to ascertain using no more than routine experimentation, many equivalents to the specific embodiments of the invention described herein. These and ail other equivalents are intended to be encompassed by the following claims.

WE CLAIM:
1. A method for automatic feature discovery for a brand object, the method
comprising the steps of :
extracting features for the brand object from a factual database; extracting features for the brand object from an opinion database; validating and mapping the features extracted from factual • database and opinion database;
generating a rank for each feature based on feature relevance and feature frequency.
2. The method as claimed in claim 1, wherein the factual database refers to a repository of information comprising objective expressions for brands objects, products ,people and their parameters.
3. The method as claimed in claim 1, wherein the step of extracting features for brand object from factual database involves queuing up the extracted features in a factual data set.
4. The method as claimed in claim 1, wherein the method comprises the steps of configuring different factual data sources;
classifying the factual data source in order of importance for performing the extraction .
5. The method as claimed in claim 1, wherein the step of extracting features
from factual database comprises controlling access to the factual data
sources.

6. The method as claimed in claim 1, wherein opinion database refers to a repository of information for brand objects comprising subjective expressions describing people's sentiments, appraisals and/or feelings towards the factual database.
7. The method as claimed in claim 1, wherein the step of extracting features from the opinion database comprises scheduling the success to these opinion data sources, is scheduled and also providing the ability to suppress access to them based on need.
8. The method as claimed in claim 1, wherein the step of validating and mapping is performed by comparing the features between factual database and opinion database.
9. The method as claimed in claim 1, wherein the step of validating and mapping uses an interpretation dictionary consisting of different parts of speech containing synonyms to the compared feature between factual database and opinion database.
10. The method as claimed in claim 1 and 9, wherein the step of validating and mapping configures relevance to each of the patterns in the interpretation dictionary for identifying relevant implicit features.
11. The method as claimed in claim 1 wherein the step of validating and mapping compares older and newer features extracted from the same data source and identifies whether the features were extracted earlier or not.
12. The method as claimed in claim 1, wherein the method comprises the steps of implicit and explicit mapping.
13. The method as claimed in claim 12, wherein the method compares by means of implicit mapping one or more factual features of one set of

factual data source against one or more features in the opinion data source.
14. The method as claimed in claim 12, wherein the implicit mapping involves a creation of a dynamic set of features which keeps growing based on mentions in different online sources.
15. The method as claimed in claim 12, wherein the implicit mapping and explicit mapping is performed as one to one, one to many, many to one and many to many
16. The method as claimed in claim 12, wherein the explicit mapping involves manual mapping of a list of unmapped entries.
17. The method as claimed in claim 12, wherein the explicit mappings are stored in implicit mapping configuration when executing implicit mapping
18. The method as claimed in claim 1, wherein the step of generating a rank for each feature based on feature relevance and feature frequency comprises sorting of feature list based on priority given in opinion database.
19. A system for automatic feature discovery for a brand object comprising :
factual data module configured to extract features for the brand object from a
factual database;
opinion data module configured to extract of features for the brand
object from an opinion database;
an execution module coupled to the factual data module and opinion data
module
configured to validate and map the features contained in factual database and opinion database and generate a rank for each feature based on feature relevance and feature frequency .

20. The system as claimed in claim 19, wherein the factual database refers to
a repository of information comprising objective expressions about objects, brands .products ,people and their parameters.
21. The system as claimed in claim 19, wherein the factual data module is configured to extract features from the brand object from factual database involving queuing up the extracted features in the factual data set.
22. The system as claimed in claim 19, wherein the factual data module is configured to different factual data sources and classifies the factual data sources in order of importance for performing the extraction.
23. The system as claimed in claim 19, wherein the unit of factual data module is configured to extract features from factual database and control access to the factual database
24. The system as claimed in claim 19„ wherein the opinion database refers to a repository of information comprising subjective expressions describing people's sentiments, appraisals or feelings towards the factual database.
25. The system as claimed in claim 19, wherein the opinion data module is configured to extract features from the opinion database and control access to the opinion database
26. The system as claimed in claim 19„wherein the execution module is configured to identify and compare features between the factual database and the opinion database using an implicit features identifier
27. The system as claimed in claim 19, wherein the execution module is configured to use an interpretation dictionary consisting of different parts

of speech containing synonyms to the compared feature between factual database and opinion database.
28. The system as claimed in claim 19, wherein the execution module is configured to allocate relevance to each pattern in the dictionary for identifying relevant implicit features.
29. The system as claimed in claim 19, wherein the execution module is configured to compares older and newer features extracted from the same data source and identify whether the features were extracted earlier or not.
30. The system as claimed in claim 19, the unit of execution module is configured to perform implicit and explicit mapping.
31. The system as claimed in claim 30, wherein the execution module is configured to compare one or more factual features of one set of factual database with one or more features in the opinion database using an 'implicit mapper'.
32. The system as claimed in claim 30 wherein the implicit mapper creates a dynamic set of features which keeps growing based on mentions in different online sources.
33. The system as claimed in claim 30, wherein the implicit mapping and explicit mapping is performed as one to one, one to many, many to one and many to many
34. The system as claimed in claim 30, wherein the explicit mapping involves a manual mapping of a list of unmapped entries.

35. The system as claimed in claim 30 wherein the explicit mappings are stored
in implicit mapping configuration when executing implicit mapping.
36. The system as claimed in claim 19, wherein the unit of execution module is
configured to sort feature list based on priority given in opinion database.