基于在线评论情感分析的社交媒体用户推荐

'Knowing your user' lies in the heart of recommendation system. However, the extant researches are weak in perceiving users’ sentiments, making it hard to identify users’ preferences, and thus affect the precision of recommendation..Therefore, in a new perspective of sentiment analysis, guided by design science and empirical study, this research proposes a social recommendation method for recommending users of social media, following the three steps of user’s feature analysis, user profiling and user recommendation. At first, through behavior study, a three-dimension feature of a user will be defined including user’s sentiments, statistics and relations, the interrelationship between the sentiments and the relations of a user will be discussed, and certain analysis will be made on that sentiment analysis serves as the foundation of quantifying user’s feature. Based on the empirical study, sentiment analysis will be applied to identify users’ preferences from online reviews in a fine-grained manner, by extracting the ‘pair of object and opinion’, and calculating the importance and the strength of user’s sentiments. And then, based on the fined-grained sentiment analysis, social network analysis will be implemented to identify opinion leader and quantify users’ social relations. After that, combining users’ preferences and social relations, a user profiling model will be proposed and the sentiment influence in social media will be calculated, and hence the prediction of potential friends will be provided for individuals, and the recommendation of target consumers will be provided for companies. In order to verify the efficiency and effectiveness of the proposed method, users’ online reviews and social relations in social media will be collected to conduct comparative experiments..This research will enrich the studies on social recommendation theoretically, and help to enhance users’ stickiness for social media, as well as implement precision marketing for business practically.

了解用户是推荐系统的核心问题。然而，现有方法缺乏对用户情感的感知，难以准确识别用户偏好，从而影响推荐精度。.为此，基于情感分析的视角，沿着“用户特征分析->用户模型构建->用户推荐”主线，拟采用设计科学与实证研究相结合的方法，实现社交媒体用户推荐。在理论层面上，通过行为研究，定义用户的情感、统计和关系三维特征，明确情感分析为特征量化的基础，分析情感与关系的相互作用。在技术层面上，通过在线评论的情感分析，提取评论对象及评价观点，计算情感的权重和强度，实现用户偏好的识别；在此基础上，进行社会网络分析，识别意见领袖，量化社会关系；结合偏好和关系，构建用户模型，量化情感影响力，从而为个人实现关注对象的预测，为企业提供目标客户的推荐。最后，以社交媒体用户的评论和关系为对象，设计对比实验，验证研究结果的有效性。.理论上，丰富社会化推荐的研究体系。实践上，提高社交媒体的用户黏性，并指导商家实施精准营销。

冷启动和数据稀疏性一直以来都是影响推荐精度的核心问题。对用户情感的感知有助于更准确地识别用户偏好，丰富用户数据，从而提高推荐的准确性。. 为此，以在线评论为研究数据，基于情感分析的视角，综合设计科学与实证研究两种研究范式，沿着“用户特征分析->用户模型构建->用户推荐模型”主线，实现社交媒体用户推荐。在此基础上，对以下内容展开。. （1）采用Python语言编写多线程爬虫程序，结合深度优先和广度优先的搜索算法，对电商平台的商品信息、在线评论以及用户数据进行采集和预处理，并经人工标注，形成实验语料。. （2）从技术层面，采用LDA主题模型和本体建模等方法，从评论中提取产品特征及用户观点，判断情感类型，并进行跨领域的鲁棒性检测，实现细粒度的用户情感挖掘；在此基础上，通过比较观点挖掘，形成产品对比关系网，从而识别用户偏好。. （3）采用计量经济方法，构建在线评论对商家业绩的影响模型，解释产品特征及评论属性对销量的影响。. （4）从行为科学角度，以用户情感为基础，结合社会网络分析技术，量化用户的社会关系和影响力，识别关键意见领袖。. （5）综合上述研究，采用深度学习、图论模型等计算框架，构建用户推荐模型，实现用户情感驱动的个性化推荐。最后，以社交媒体用户为对象，设计对比实验，验证研究结果的有效性。. 理论上，丰富社会化推荐的研究体系。实践上，提高社交媒体的用户黏性，并指导商家实施精准营销。