基于用户评论情感分析的O2O平台推荐策略研究

Research on Recommendation Strategy of O2O Platform Based on Sentiment Analysis of User Reviews

传统推荐策略基于价格、销量、评分、地理位置等结构化数据,而未能利用具有丰富偏好信息的用户评论文本数据。基于用户评论的情感词构建推荐策略,将用户评论数据转化为比用户购买和评分更加深刻和全面的用户偏好、商户特征,实现更加个性化的推荐,节省用户的信息加工成本,提高订单的转化率,为盘活海量数据、提升数据价值提供新思路。本文通过深度挖掘用户评论文本信息,构建基于多维度用户偏好和商户特征模型的个性化推荐策略。以大众点评网为例,选取特定用户和商户,采集其评论文本数据,将用户关注权重和商户多维属性的情感分析评分计算建立数学模型,得出推荐指数。通过问卷调查验证推荐策略的有效性,显示有85.0%的受访者在阅读用户评论后的选择和自动推荐指数前三位商户达成一致。

[Goal/Significance]Traditional recommendation strategies are based on structured data such as price,sales,ratings,geographic location,etc.,and fail to take advantage of user review text data with rich preference information.Constructing recommendation strategies based on the emotion words of user comments transforms user comment data into more profound and comprehensive user preferences and merchant characteristics than user purchases and ratings to achieve more personalized recommendations,save users’information processing costs,improve the conversion rate of orders,and provide a new way of thinking to revitalize the huge amount of data and enhance the value of data.[Methods/Process]This paper constructs a personalized recommendation strategy based on multi-dimensional user preference and merchant characteristics model by deeply mining user review text information.Taking Dianping.com as an example,specific users and merchants are selected,their comment text data are collected,and the mathematical model is established by calculating the user attention weights and the sentiment analysis scores of multi-dimensional attributes of the merchants to derive the recommendation index.[Results/Findings]The effectiveness of the recommendation strategy is verified through a questionnaire survey,which shows that 85.0%of the respondents agree with the top three merchants of the automatic recommendation index in their choices after reading user reviews.