Application Of Consumer's Comment Mining In Restaurant Recommendation System

The rapid development of electronic commerce makes online consumption an indispensable part of people's life.Recommendation system helps consumers make better decisions through higher priority.Traditionally,restaurant recommendation system only takes consumers' overall scores into consideration.However,different characteristics of the restaurant,including food taste,decoration,equipment,atmosphere and service,are ignored in rating the restaurants on a scale of zero to five.This may lower the accuracy of its scoring system and make people turning to read consumers' comments.These comments contain several valuable information,such as suggestions of rational consumption,dining experiences and feelings toward restaurant services.These information can help to portray features of the restaurant accurately and improve the accuracy of recommendation system.This paper uses natural language processing technology to analyze consumer's comments and apply them to restaurant recommendation system.Moreover,this study aims at data mining,especially for the consumption suggestions,feelings and eventually,improve the accuracy of restaurant recommendation system.The main contents of this paper are as follows:(1)This study constructs a restaurant recommendation sub-system based on consumer's textcomments mining.Six modules are included as basis: data collection;data pre-management;textbased summary and extraction;sentimental analysis;similarity calculation;restaurant recommendation.These modules,which constitute a complete off-line recommendation algorithm and system,realize intelligent functions including data collection,data pre-management,text-based summary and extraction,sentimental analysis,similarity calculation,restaurant recommendation.(2)This study uses BeautifulSoup base to acquire more than two million data information from catering sites,including user IDs,restaurant IDs,consumer's scores and text comments.These data information are stored in MongoDB database.(3)The study applies dependency syntactic analysis technology to text-based summary and extraction module.After pre-management,text-comments are subjected to the dependency syntactic analysis technology so as to obtain grammatical relationships between words.Then,the study gather the words which match the logic of natural language and generate a complete description as critical abstraction.Meanwhile,marks the corresponding comment sentences as subject sentences.This method makes full use of the grammatical relations between words and has more advantages than existing methods.(4)The study applies part-of-speech feature selection model to sentimental analysis module.First of all,artificially selects high and meaningful part-of-speech combinations in statistical magnitude by analyzing different combinations in subjective sentences.Then,by combining features of phrases selection,the study gathers the phrases which suit the condition of particular part-of-speech combination as the based set.Finally,based on several commendatory subjective sentences and derogatory subjective sentences marked by human beings,the sub-system is trained by logistic regression in order to distinguish different emotions.This method has a certain improvement in classification performance over the traditional methods.(5)Instead of using the traditional technology,the study constructs a multi-attribute scoring system and calculate the similarity of restaurants by combining the results of text-based extraction module and sentimental analysis module.The new method adopted by this system portrays different characteristics of each restaurant,and has higher accuracy of recommendation results over the traditional recommendation system.