基于评论文本情感分析和概率模型的汽车推荐系统研究

发布时间：2018-05-28 15:39

本文选题：汽车推荐系统 + 文本分析　；参考：《重庆大学》2016年硕士论文

【摘要】：伴随着国家大数据战略的实施,以电子商务为首的互联网应用与现代生活深度融合的同时,也逐渐促进了汽车等传统行业市场经营和发展模式的转型。互联网平台和信息技术的发展为消费者选购汽车提供了更多的便利条件和参考信息来源,然而汽车销售市场愈来愈激烈的竞争却导致了信息负载问题的进一步加剧。推荐系统作为一种主动过滤无关信息的技术,特别是以基于内容和协同过滤的改良算法为主流的推荐技术,凭借其在普通消费商品推荐任务中所展现的出色的性能和效率,不仅在电子商务平台、互联网金融、社交网络等各大互联网平台中得到了广泛应用,同时也成为大量学术领域研究者关注的热点。然而对于一些特定领域场景下的个性化推荐任务,例如房屋、汽车、金融产品等,协同过滤技术和以物品过滤为基础的推荐算法所表现出的有效性并不能很好地满足用户需求。本文在针对传统推荐算法在汽车个性化推荐任务中所遇到的困难和瓶颈进行了深入分析的基础上,构建了一种基于概率模型的汽车个性化推荐系统APRS。该推荐系统可以帮助普通消费者在无需过多了解汽车领域相关知识的情况下,辅助他们进行汽车的挑选以及购买决策。论文的主要研究成果如下:(1)针对中文商品在线评论的特点,本文在挖掘观点分句Part-Of-Speech(POS)模式和构建相关专业领域知识词典的基础上,提出了一种基于商品属性特征的意见挖掘和情感分析算法SAOSP,该算法以评论分句为粒度对商品在线评论文本进行分析,从而完成了商品各项属性特征评分情感分析和量化工作,从客观数据事实的角度对商品基本信息进行补充的同时,也为推荐模型的构建奠定了数据基础。(2)通过对用户在汽车购买过程中的行为特征进行综合分析,我们在论文中提出了一种新颖的概率模型用于对用户的商品购买过程进行描述。本文在该模型中综合考虑了用户个人偏好、互联网用户群体以及现实中社交群体等影响用户最终购买决策的各项因素,并利用矩阵投影算法对模型中各概率参数进行计算。(3)针对特定应用领域商品复杂度高,普通用户相关专业领域知识匮乏的特点,本文所提出的推荐系统采用会话的方式对用户的用途需求进行显式引导,基于所提出的推荐模型将用户的用途需求与商品属性特征信息之间潜在的联系进行构建和量化,并给出了Top-K推荐列表中商品评分排名的计算公式。帮助普通用户在缺少汽车相关专业领域知识的情况下,只需根据自身的购车用途需求可以得到基于客观数据事实的推荐结果。(4)通过引入FMM和TR两种基于协同过滤技术的推荐算法作为对比,本文在现实数据集上对所提出的汽车个性化推荐系统APRS进行了对比实验分析。实验结果验证,APRS在汽车这种特定应用领域的商品的个性化推荐任务中具有较高的有效性和准确性。
[Abstract]:With the implementation of the national big data strategy, the Internet application led by electronic commerce and the deep integration of modern life have also gradually promoted the transformation of the traditional industry market management and development model. The development of the Internet platform and information technology provides more conveniences and reference letters for consumers to buy cars. Source, however, the increasingly fierce competition in the car market has led to a further exacerbation of the problem of information load. The recommendation system is a technology that is active in filtering unrelated information, especially the recommendation technology based on content based and collaborative filtering, which is presented in the recommendation task of common consumer goods. Excellent performance and efficiency have been widely used not only in the Internet platforms, such as e-commerce platform, Internet finance, social network and so on, but also the focus of researchers in a large number of academic fields. However, personalized recommendation tasks, such as houses, cars, financial products, etc., are coordinated in some specific field scenarios. The effectiveness of the filtering technology and the recommendation algorithm based on the object filtering can not satisfy the user's requirements well. Based on the analysis of the difficulties and bottlenecks of the traditional recommendation algorithm in the personalized recommendation task of the automobile, this paper constructs a personalized recommendation of automobile based on the probability model. The main research results of the paper are as follows: (1) in view of the characteristics of the online comment on Chinese goods, this article is exploring the Part-Of-Speech (POS) model and structure of the point of view. (1) the main research results of this paper are as follows: (1) in view of the characteristics of the online comment on Chinese goods, this article is mining the model and structure of the point clause (POS). On the basis of the knowledge dictionary of related professional fields, a kind of opinion mining and emotional analysis algorithm SAOSP based on the characteristics of the commodity is proposed. The algorithm uses commentary clause as the granularity to analyze the online comment text of the commodity, thus completing the analysis and quantification of the commodity attribute characteristic score feeling and the quantitative work, from the objective data facts. While adding the basic information to the commodity, it also lays a data base for the construction of the recommended model. (2) through a comprehensive analysis of the behavior characteristics of the user in the process of automobile purchase, we put forward a novel probability model in this paper to describe the process of the purchase of the user. This paper is in this model. The factors that affect user's final purchase decision are comprehensively considered, such as user preferences, Internet user groups and social groups in reality. And the matrix projection algorithm is used to calculate the probability parameters in the model. (3) the characteristics of the high complexity of the commodity in the specific application field and the lack of knowledge in the related professional fields of the ordinary users, The proposed system uses a conversational approach to explicitly guide the user's use requirements. Based on the proposed model, it constructs and quantifies the potential connections between the user's use requirements and the feature information of the commodity, and gives the calculation formula of the ranking of the commercial products in the Top-K recommendation list. In the absence of automobile related professional knowledge, we only need to get the results based on objective data facts according to their own car use requirements. (4) by introducing two recommended algorithms based on FMM and TR as comparison, the proposed auto personalized recommendation system APRS is presented on the real data set. The experimental results show that APRS has high effectiveness and accuracy in the personalized recommendation task of the goods in this particular application field.
【学位授予单位】：重庆大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP391.3

【相似文献】