基于多粒度犹豫模糊语言信息的群推荐方法研究

发布时间：2018-01-31 16:34

本文关键词： 在线社交网络多粒度犹豫模糊语言群偏好获取方法群推荐方法　出处：《合肥工业大学》2017年博士论文　论文类型：学位论文

【摘要】：群推荐系统是为参与共同活动的群体推荐满足群中所有成员的共同爱好的信息系统,现有成果主要支持线下已成团群体。随着在线社交网络的出现,具有相似兴趣的群体形成了各类虚拟社区,面向社交网络中虚拟社区的群推荐系统将成为推荐领域的热点问题。由于群推荐方法是群推荐系统的核心问题,因此研究群推荐理论与群推荐方法具有重要的理论意义和应用价值。在群推荐系统中,由于不同的个体偏好不同,个体在描述这些偏好信息时习惯采用自然语言,所以其偏好信息通常具有模糊性、犹豫性的特点。此外,由于不同的网络平台上可能采用不同粒度的语言信息进行推荐,所以群体偏好信息还具有多粒度性。基于此,本文在犹豫模糊语言环境下,首先研究了在线社交网络中的用户群发现方法。其次,在研究了个体评分预测的基础上,深入地研究了在线社交网络中用户群偏好获取方法。最后,研究了面向在线社交网络用户的TOPSIS群推荐方法、VIKOR群推荐方法。具体研究工作与创新点如下:(1)多粒度犹豫模糊语言环境下的群体发现方法。群体发现是群推荐的基础性问题,为此本部分首先在多粒度犹豫模糊语言术语集的基础上,引入了多粒度犹豫模糊语言余弦相似性计算公式,采用余弦相似性公式计算用户之间的相似性,分析余弦相似性与距离相似性公式之间的差异。其次,将最小生成树方法拓展到多粒度犹豫模糊语言环境中用于聚类分析。最后,将最小生成树聚类方法与等价关系聚类方法进行对比,分析所提出的方法聚类分析中的合理性和有效性。(2)犹豫模糊语言环境下的评分预测方法。评分预测问题是群推荐系统中研究焦点之一,为此本部分首先综述群推荐系统中评分预测的主要方法,阐述犹豫模糊语言环境下评分预测的必要性。其次,在犹豫模糊语言环境下距离相似性公式、余弦相似性公式的基础上提出相关系数相似性公式用于计算用户之间的相似性。最后,采用距离相似性公式、余弦相似性公式和相关系数相似性公式对算例进行计算,预测出未知的评分信息,比较三种方法预测的精度,分析所提出的相关系数相似性公式对犹豫模糊语言信息评分预测的可行性和有效性。(3)多粒度犹豫模糊语言的群偏好获取方法。在群推荐系统中如何将个体偏好信息集结成群体偏好信息是一个关键问题,为此本部分首先提出三角形犹豫模糊集的概念,分析三角形犹豫模糊集的性质,采用三角形犹豫模糊集对多粒度犹豫语言进行转换。其次,定义广义三角形犹豫模糊有权平均算子和广义三角形犹豫模糊有权几何算子,推导出这两个算子的性质。最后以汽车推荐为例,利用这两个算子对多粒度犹豫模糊语言信息描述的群偏好问题进行集结,分析该模型解决群体偏好获取是合理和有效的。(4)多粒度犹豫模糊语言的TOPSIS群推荐方法。针对不同群体的偏好信息具有多粒度性、犹豫模糊性等特点,本文首先定义多粒度犹豫模糊语言术语集的概念,定义多粒度犹豫模糊语言的系列距离公式,研究这些公式的性质,讨论公式之间的关系。其次,在属性权重完全未知的情况下,建立目标规划模型,利用拉格朗日方程求解模型得到属性权重;在属性权重不完全未知的情况下,采用线性规划模型求解属性权重。最后,将这些距离公式结合TOPSIS方法用于群体推荐问题,并分析公式的参数对TOPSIS方法的满意度及推荐结果的影响情况。(5)多粒度犹豫模糊语言信息的VIKOR群推荐方法。针对群推荐系统中被推荐项目具有多粒度性、犹豫模糊性问题,本文首先在多粒度犹豫模糊语言术语集的基础上,引入多粒度犹豫模糊语言信息熵的概念及计算公式,采用信息熵公式计算被推荐项目的属性权重;其次,将传统的VIKOR方法拓展到多粒度犹豫模糊领域,并对其妥协解公式进行改进,将改进的VIKOR方法用于群推荐;最后,从理论分析、数值计算和敏感性分析3个方面将VIKOR方法与TOPSIS方法进行对比,分析所提出的方法在群推荐应用中的合理性和有效性。总之,本文在多粒度犹豫模糊语言环境下,研究了在线社交网络用户的群发现方法、用户群偏好获取方法和群推荐方法,对群推荐系统中的关键问题进行了深入、系统的研究,研究成果不仅拓展了模糊数学理论,同时对群推荐等群决策问题具有指导意义。
[Abstract]:Group recommendation system is the common information recommendation system of all members of the group in order to meet to participate in the common activities of the group, the existing main achievements under the support line has become a group of groups. With the emergence of online social networks, with similar interest groups formed of various types of virtual communities, to group recommendation system of virtual community in social network will has become a hot issue in the field of group recommendation. The recommended method is a core problem in group recommendation system, so it has important theoretical significance and application value to research the theory and method of group recommendation group recommendation. In group recommendation system, due to different individual preferences of different individual habits, the use of natural language in describing the preference information, so the the preference information usually is fuzzy, the characteristics of hesitation. In addition, due to the different network platform may recommend the use of language information with different granularity, so The preference information also has multi granularity. Based on this, this article in hesitant fuzzy language environment, firstly, users found in online social networks. Secondly, based on the individual rating prediction, in-depth study of the preference of users in online social network acquisition method. Finally, the research oriented online social network user group TOPSIS recommendation method, recommended VIKOR group. The specific research work and innovations are as follows: (1) multi granularity hesitant fuzzy language environment. The group found that the group found that the method is recommended the fundamental problem, so this part of the multi granularity based hesitation fuzzy linguistic terms set. The introduction of multi granularity hesitant fuzzy language cosine similarity formula. The similarities between users using cosine calculation formula, analysis between similarity and cosine distance similarity formula Difference. Secondly, minimum spanning tree method is extended to multi granularity clustering analysis for hesitant fuzzy language environment. Finally, comparing the relationship between the minimum spanning tree clustering method and clustering method, clustering analysis method proposed in this paper is reasonable and effective. (2) prediction method for hesitant fuzzy language environment score the problem is the focus of research. The prediction score of group recommendation system, the main method of this part of the first group recommendation system score, necessity of hesitant fuzzy language environment prediction. Secondly, fuzzy language under the environment of distance similarity formula in hesitation, based on the cosine similarity formula proposed the correlation coefficient of similarity the formula for computing the similarity between users. Finally, the similarity distance formula, cosine similarity formula and the correlation coefficient of similarity calculation formula The calculation, predict the unknown information prediction score, compare the three methods of precision, correlation coefficient analysis the feasibility of similar formula for predicting hesitant fuzzy language information and effectiveness. (3) multi granularity hesitation group preference fuzzy language acquisition method. In the group recommendation system to individual preference information aggregated into group preference information is a key problem, this part first proposes the concept of triangle hesitant fuzzy sets, analysis of properties of triangle hesitant fuzzy set using triangular fuzzy sets to hesitate, hesitate to convert multi granularity language. Secondly, the definition of generalized triangle fuzzy weighted average operator and hesitation hesitant fuzzy generalized triangle right geometric operator nature, the two operators are derived. Finally, to recommend car as an example, using the two operator fuzzy language description hesitate to ask for multi granularity partial group Questions by aggregating group preference acquisition analysis solution is reasonable and effective. The model (4) multi granularity hesitate method recommended fuzzy language TOPSIS group. With multi granularity according to different groups of preference information, hesitant fuzzy characteristics, the paper firstly defines the concept of multi granularity fuzzy linguistic terms set hesitation, defined the size of the distance formula of fuzzy language series of hesitation, properties of these formulas, discussed the relationship between the formula. Secondly, the attribute weights are completely unknown, the establishment of multi-objective programming model, using Lagrange equation model to obtain attribute weights; in the incomplete attribute weights are unknown, using linear programming models to compute the weights of attributes. Finally, the distance formula combined with TOPSIS method for group recommendation problem, and to analyze the impact of satisfaction and recommendation results the parameters of the formula of TOPSIS method. . (5) multi granularity fuzzy linguistic information recommendation method. VIKOR group. In group recommendation system recommended by the project with multi granularity, hesitant fuzzy problem, firstly, hesitate in the basis of fuzzy multi granularity linguistic term set, introducing multi granularity hesitate concept and formula of fuzzy language information entropy, using the formula the information entropy calculation recommended attribute weights of the project; secondly, the traditional VIKOR method is extended to multi granularity fuzzy field and the hesitation, the compromise solution formula was improved, the improved VIKOR method for group recommendation; finally, from the theoretical analysis, numerical calculation and sensitivity analysis of the 3 aspects of the VIKOR method and TOPSIS method comparative analysis method, proposed in the group recommended the application of rationality and validity. In a word, based on the multi granularity hesitant fuzzy language environment, the online social network user group is Method, user group preference acquisition method and group recommendation method, have carried out in-depth and systematic research on the key issues of group recommendation system. The research results not only expand the fuzzy mathematics theory, but also have guiding significance for group recommendation and other group decision making problems.

【学位授予单位】：合肥工业大学
【学位级别】：博士
【学位授予年份】：2017
【分类号】：TP391.3

【参考文献】