个性化搜索引擎推荐算法研究
发布时间：2019-06-20 22:19
【摘要】： 随着Internet和网络信息技术的迅猛发展,网络资源呈指数急剧增长,传统的通用搜索引擎的查询结果只依赖于查询关键词,而实际上,即便相同的查询词,不同的用户查询目的可能不同,所希望的返回结果也会因人而异。针对这种情况,人们迫切需要一种针对个人特点提供更加精确查询结果的搜索工具,以用户为中心的个性化搜索引擎便应运而生。 本文首先全面了解了实现个性化搜索引擎的基本理论和研究现状,并对现有各种个性化推荐技术进行性能对比分析,为以后的研究提供了理论基础。 接着,本文研究了推荐领域最重要的协同过滤算法,基于用户推荐的协同过滤可以为用户发现新的潜在感兴趣的资源,但是具有稀疏性等缺点;基于项目推荐的协同过滤在某种程度上可以解决稀疏性,而且简单有效,但是只能发现和用户已有兴趣相似的信息。针对这些问题,本文提出了一种基于单值分解的集影响协作过滤推荐算法,利用单值分解和增大影响集来提高协同过滤的推荐质量,解决稀疏性问题,改善推荐系统的性能。 然而在应用了改进的协同过滤推荐算法的推荐系统中,除了已经解决的稀疏性问题,还存在着冷开始新项目问题、扩展性问题以及用户潜在兴趣难以挖掘等,本文在前面研究的基础上,提出了一种个性化推荐融合算法,在优秀的基于用户协同过滤推荐思想基础上,结合现有矩阵技术,扩展影响集,利用基于项目协同过滤以及基于内容过滤,解决了稀疏问题、扩展性问题、冷开始和用户潜在兴趣难以挖掘等问题,提高了推荐系统的推荐质量。并在此基础上,提出了一种策略预测用户评分,解决了由于用户对资源苛刻程度不同,而导致评分相差较大的问题。 最后,分析研究了开源全文检索工具Lucene,并在该平台上加入了个性化搜索模块,分别对改进的协作过滤推荐算法和个性化推荐融合算法进行了仿真实验。实验结果表明:改进的协作过滤推荐算法比传统的协同过滤算法的推荐质量高,而在冷开始状况下,个性化推荐融合算法比改进的协作过滤推荐算法推荐质量高,预测评分更加与实际评分相接近,搜索结果更加符合用户需求,提高了个性化搜索引擎的服务质量。
[Abstract]:With the rapid development of Internet and network information technology, the network resources increase rapidly. The query results of the traditional general search engine only rely on query keywords. In fact, even if the same query words, different user query purposes may be different, the desired return results will vary from person to person. In view of this situation, people urgently need a search tool to provide more accurate query results according to personal characteristics, and a user-centered personalized search engine emerges as the times require. In this paper, the basic theory and research status of personalized search engine are fully understood, and the performance of various personalized recommendation technologies is compared and analyzed, which provides a theoretical basis for future research. Then, this paper studies the most important collaborative filtering algorithm in the field of recommendation. Collaborative filtering based on user recommendation can find new potentially interested resources for users, but it has some shortcomings, such as sparsity. Collaborative filtering based on project recommendation can solve sparsity to some extent, and is simple and effective, but can only find information similar to the interest of users. In order to solve these problems, a set-influence cooperative filtering recommendation algorithm based on single-valued decomposition is proposed in this paper. Single-valued decomposition and increasing influence set are used to improve the recommendation quality of collaborative filtering, solve the sparsity problem and improve the performance of recommendation system. However, in the recommendation system which applies the improved collaborative filtering recommendation algorithm, in addition to the sparsity problem that has been solved, there are also cold start new project problems, expansibility problems and difficult to mine the potential interest of users. On the basis of the previous research, this paper proposes a personalized recommendation fusion algorithm. On the basis of the excellent recommendation idea based on user collaborative filtering, combined with the existing matrix technology, the impact set is extended. By using project-based collaborative filtering and content-based filtering, the sparse problem, expansibility problem, cold start and difficult to mine the potential interest of users are solved, and the recommendation quality of recommendation system is improved. On this basis, a strategy is proposed to predict the user score, which solves the problem that the score is different because the user is harsh on the resource. Finally, the open source full-text retrieval tool Lucene, is analyzed and studied, and the personalized search module is added to the platform, and the improved collaborative filtering recommendation algorithm and personalized recommendation fusion algorithm are simulated respectively. The experimental results show that the recommendation quality of the improved collaborative filtering recommendation algorithm is higher than that of the traditional collaborative filtering algorithm, but at the beginning of cold, the personalized recommendation fusion algorithm has higher recommendation quality than the improved collaborative filtering recommendation algorithm, the prediction score is more close to the actual score, the search results are more in line with the needs of users, and the quality of service of personalized search engine is improved.
【学位授予单位】：湖南大学
【学位级别】：硕士
【学位授予年份】：2009
【分类号】：TP391.3
本文编号：2503553
[Abstract]:With the rapid development of Internet and network information technology, the network resources increase rapidly. The query results of the traditional general search engine only rely on query keywords. In fact, even if the same query words, different user query purposes may be different, the desired return results will vary from person to person. In view of this situation, people urgently need a search tool to provide more accurate query results according to personal characteristics, and a user-centered personalized search engine emerges as the times require. In this paper, the basic theory and research status of personalized search engine are fully understood, and the performance of various personalized recommendation technologies is compared and analyzed, which provides a theoretical basis for future research. Then, this paper studies the most important collaborative filtering algorithm in the field of recommendation. Collaborative filtering based on user recommendation can find new potentially interested resources for users, but it has some shortcomings, such as sparsity. Collaborative filtering based on project recommendation can solve sparsity to some extent, and is simple and effective, but can only find information similar to the interest of users. In order to solve these problems, a set-influence cooperative filtering recommendation algorithm based on single-valued decomposition is proposed in this paper. Single-valued decomposition and increasing influence set are used to improve the recommendation quality of collaborative filtering, solve the sparsity problem and improve the performance of recommendation system. However, in the recommendation system which applies the improved collaborative filtering recommendation algorithm, in addition to the sparsity problem that has been solved, there are also cold start new project problems, expansibility problems and difficult to mine the potential interest of users. On the basis of the previous research, this paper proposes a personalized recommendation fusion algorithm. On the basis of the excellent recommendation idea based on user collaborative filtering, combined with the existing matrix technology, the impact set is extended. By using project-based collaborative filtering and content-based filtering, the sparse problem, expansibility problem, cold start and difficult to mine the potential interest of users are solved, and the recommendation quality of recommendation system is improved. On this basis, a strategy is proposed to predict the user score, which solves the problem that the score is different because the user is harsh on the resource. Finally, the open source full-text retrieval tool Lucene, is analyzed and studied, and the personalized search module is added to the platform, and the improved collaborative filtering recommendation algorithm and personalized recommendation fusion algorithm are simulated respectively. The experimental results show that the recommendation quality of the improved collaborative filtering recommendation algorithm is higher than that of the traditional collaborative filtering algorithm, but at the beginning of cold, the personalized recommendation fusion algorithm has higher recommendation quality than the improved collaborative filtering recommendation algorithm, the prediction score is more close to the actual score, the search results are more in line with the needs of users, and the quality of service of personalized search engine is improved.
【学位授予单位】：湖南大学
【学位级别】：硕士
【学位授予年份】：2009
【分类号】：TP391.3
【引证文献】
相关期刊论文 前1条
1 袁莉萍;;专业化音乐教学信息搜索引擎技术研究[J];现代计算机(专业版);2011年05期
相关硕士学位论文 前4条
1 白晓波;基于事件驱动模型的搜索引擎的研究及原型系统设计[D];湖南大学;2010年
2 代旭峰;基于用户兴趣模型的搜索引擎结果推荐系统[D];复旦大学;2011年
3 谭明辉;基于web数据挖掘的个性化搜索引擎的研究与应用[D];江西农业大学;2012年
4 文新胜;基于专家池的协同过滤推荐系统研究[D];湖北大学;2011年
，本文编号：2503553
本文链接：https://www.wllwen.com/kejilunwen/sousuoyinqinglunwen/2503553.html
