Sentiment extraction from consumer reviews for providing product recommendations

Abstract

A system and method for recommending a product to a user in response to a query for a product with a feature wherein the recommendation is accompanied by a quotation expressing a sentiment about the feature or the product.

1 .- 21 . (canceled) 
     
     
         22 . A recommendation system comprising a memory storing computer readable program code for recommending a product to a user, wherein the recommendation system is configured to execute the computer readable program code to:
 receive a search query from a user device associated with the user;   determine a product feature of the search query;   determine a product corresponding to the product feature, wherein determining the product corresponding to the product feature comprises applying a semantic template to the search query, and wherein the semantic template is applied to the search query by a span query;   retrieve a product profile corresponding to the product;   retrieve a quotation associated with the product and the product feature, wherein the quotation comprises expression of a sentiment about the product or the product feature; and   output a product recommendation comprising the product profile and the quotation to the user device.   
     
     
         23 . The recommendation system of  claim 22 , wherein the recommendation system comprises a recommendation database and applying the semantic template to the search query comprises:
 building the semantic template from the search query, the semantic template comprising atomic semantic templates corresponding to the product feature; and   searching the recommendation database with the semantic template.   
     
     
         24 . The recommendation system of  claim 23 , wherein a plurality of product features, a plurality of sentiments, a plurality of quotations, and a plurality of sentiment volatilities are stored in the recommendation database. 
     
     
         25 . The recommendation system of  claim 24 , wherein the recommendation system is further configured to execute the computer readable program code to determine a score for each product feature of the plurality of product features stored in the recommendation database. 
     
     
         26 . The recommendation system of  claim 23 , wherein the atomic semantic templates are an unordered set of atomic semantic templates. 
     
     
         27 . The recommendation system of  claim 23 , wherein determining each score comprises using the plurality of sentiments, wherein the plurality of sentiments are extracted from a plurality of documents. 
     
     
         28 . The recommendation system of  claim 27 , wherein at least one of the plurality of documents is a web page of a manufacturer website or retailer website, the web page including a description of the product. 
     
     
         29 . The recommendation system of  claim 28 , wherein the web page further includes one or more consumer comments about the product. 
     
     
         30 . The recommendation system of  claim 27 , wherein the recommendation system is further configured to execute the computer readable program code to determine an overall sentiment for each document of the plurality of documents. 
     
     
         31 . The recommendation system of  claim 30 , wherein the recommendation system is further configured to execute the computer readable program code to determine a sentiment volatility of the overall sentiment for each document of the plurality of documents. 
     
     
         32 . The recommendation system of  claim 31 , wherein the recommendation system is further configured to execute the computer readable program code to determine a reliability score for each document of the plurality of documents. 
     
     
         33 . The recommendation system of  claim 32 , wherein the reliability score for each document is based on the sentiment volatility of each document. 
     
     
         34 . The recommendation system of  claim 23 , wherein building the semantic template from the search query comprises:
 extracting one or more product features from the search query; and   sorting the extracted one or more product features based on a lexicography.   
     
     
         35 . The recommendation system of  claim 22 , wherein outputting the product recommendation comprises:
 sorting a plurality of product profiles and quotations returned by the search query; and   identifying the product recommendation from the sorted search product profiles and quotations.   
     
     
         36 . The recommendation system of  claim 35 , wherein the plurality of product profiles and quotations are sorted based on product feature scores. 
     
     
         37 . The recommendation system of  claim 22 , wherein the search query is received from the user device, via a network, by a front end server of the recommendation system. 
     
     
         38 . The recommendation system of  claim 22 , wherein the product profile comprises a product name and one or more product ratings associated with the product. 
     
     
         39 . The recommendation system of  claim 22 , wherein the quotation is extracted from a previously received document associated with the product. 
     
     
         40 . The recommendation system of  claim 22 , wherein the product feature comprises an item characteristic of a product. 
     
     
         41 . The recommendation system of  claim 22 , wherein the product feature comprises an abstract characteristic of a product.