摘 要

21世纪全球已经步入以信息化为核心，一轮又一轮的信息革命浪潮翻涌而来，人工智能、物联网、云计算等相关领域的研究与应用大规模深入展开。时下新兴科学研究发展速度之快，发展规模之大、影响范围之广，使得在现代信息技术大展拳脚的同时，也伴随着呈爆炸式增长态势的海量信息产生。由这样的背景下，引发了一系列“信息过载”问题，使得互联网用户在获取有效信息上所耗费的时间与精力大大增加，不仅严重影响用户的使用体验，甚至有可能由于无效或错误信息而造成不可挽回的损失。因此如何科学合理地为用户推荐其感兴趣或有需求的信息，筛选过滤掉无意义或者不符合用户期望的信息成为目前相关领域研究的一大难点。

信息过载问题的处理工具包括搜索引擎、推荐系统等，其中推荐系统通过挖掘用户历史评分信息，利用机器学习等方法构建推荐模型，实现对用户未知项目评分的预测、推荐结果的生成。目前推荐系统已被广泛应用于多方位多领域，然而大部分网络平台提供的推荐服务属于个性化推荐技术，无法满足当一个群体作为研究对象时的系统需求。已有研究通过引入社交网络机制提升推荐系统性能，但仍然存在着一些不足之处，本文改进基于信任网络的个体推荐模型，并引入共识演化网络机制实现群体推荐模型的构建。

本文的具体研究工作将从以下四个方面展开：

1. 首先阐明文章的研究背景及意义，并对国内外关于群体推荐系统领域的研究现状进行分析；简要介绍目前广泛应用的几类推荐算法，其中详细说明了基于用户的协同过滤算法机制；然后对信任网络的相关概念进行阐述，为下一步改进个体推荐模型提供理论支撑。
2. 在传统基于用户的协同过滤算法基础上，引入信任网络机制。首先综合多角度度量用户评分可靠程度，其次在信任传递计算模型中提出传播效率因子，并引入Hamacher乘积算子、有序加权平均算子以提高信任关系计算准确性，基于信任关系构建个体推荐模型预测评分，并为下一步群体推荐模型提供数据来源。
3. 将共识演化网络机制引入上述个体推荐模型，通过一系列转换计算构建群体共识演化网络，并考虑到群体内个体用户之间的话语权大小存在差异，因此以用户评分的数目作为衡量标准，提出一个新共识指数赋予用户不同的重要性，参照调整原则修正用户共识，促使群体意见或行为趋于一致，生成群体推荐结果。
4. 基于公开数据集MovieLens完成上述两个模型的测试，通过对比实验确定模型中的未知参数，实验结果表明本文提出的模型有效提升推荐系统的精准度。

关键词： 群体推荐系统；协同过滤；信任网络；共识演化网络

Abstract

In the 21st century, the world has entered the core of informatization, and round after round of waves of information revolution have surged in, Internet of Things, cloud computing and other related fields of research and application of large-scale in-depth. With the rapid development of new scientific research, the large scale of development and the wide range of influence, the modern information technology is developing rapidly. Moreover, it is accompanied by the explosion of mass information. In this thesis, a series of "information overload" problems have arisen, which greatly increase the time and energy spent by Internet users in obtaining effective information, and not only seriously affect the user experience, there is even the possibility of irreparable loss due to invalid or incorrect information. Therefore, how to scientifically and reasonably recommend the information that users are interested in or need, and filter out the information that is meaningless or does not meet the users' expectations has become a major difficulty in current research in related fields.

Tools for dealing with information overload problems include search engines and recommendation systems, in which recommendation systems build recommendation models by mining user history score information and using machine learning, to achieve the user unknown item score prediction, recommendation of the results. Recently, recommendation system has been widely used in many fields, but most of the recommendation services provided by the network platform are personalized recommendation technology, which can not meet the needs of the system when a group is used as a research object. Some studies have been done to improve the performance of recommendation system by introducing social network mechanism, but there are still some shortcomings. This thesis improves the individual recommendation model based on trust network, the consensus evolution network mechanism is introduced to construct the group recommendation model.

The specific research work of this thesis will be carried out from the following four aspects:

1. The background and significance of this thesis are explained, and the research status of Group Recommendation System at domestic and abroad is analyzed. Several kinds of recommendation algorithms are widely used at present are briefly introduced, the mechanism of user-based collaborative filtering algorithm is explained in details, and then the related concepts of trust network are explained, which provides the theoretical support for improving the individual recommendation model.
2. Based on the traditional user-based collaborative filtering algorithm, the trust network mechanism is introduced. Firstly, the reliability of user ratings is measured from multiple angles; secondly, the propagation efficiency factor is proposed in the trust transfer calculation model, and the Hamacher aggregation operator and ordered weighted average operator are introduced to improve the accuracy of trust relationship calculation, the individual recommendation model was constructed based on trust relationship to predict the score and provide the data source for the next group recommendation model.
3. A consensus evolution network mechanism is introduced into the above individual recommendation model, and a group consensus evolving network is constructed through a series of transformation calculations, taking into account the difference in the voice power of individual users within the group, based on the number of users' ratings, a new consensus index is proposed, which gives users different importance, modifies users' consensus according to the principle of adjustment, makes group opinions or behaviors tend to be consistent, and generates group recommendation results.
4. The two models are tested based on the open data set MovieLens, and the unknown parameters in the models are determined by comparative experiments. The experimental data show that the proposed model can effectively improve the accuracy of the recommender system.

Keywords：Group recommender system; collaborative filtering; trust network; consensus evolution network

目 录

第1章 绪论 1

1.1 研究背景 1

1.2 国内外研究现状 2

1.3 研究意义 3

1.4 论文组织结构 3

第2章 相关理论及技术介绍 5

2.1 推荐系统及相关技术 5

2.1.1 基于协同过滤技术的推荐概述 5

2.1.2 推荐系统的性能指标 8