兰州大学机构库 >信息科学与工程学院
Alternative TitleCollaborative Filtering Recommendation Based on Tag and Time Information
Thesis Advisor林和
Degree Grantor兰州大学
Place of Conferral兰州
Degree Name硕士
Degree Discipline计算机技术
Keyword推荐搜索 协同过滤 标签 矩阵填充 时间权重 兴趣转移
方法 ——首先,根据标签所反映的资源描述信息和用户兴趣偏好信息将其划分为标准化标签和社会化标签,并对标签相似度计算公式做出改进。其次,在传统协同过滤的基础上引入标签信息。使用改进后的标签相似度计算公式进行第一次评分值预测,并将评分预测值填充于原始评分矩阵;在填充后的矩阵上进行基于用户的最近邻推荐,得到最终的推荐结果。最后,针对用户兴趣转移问题对融合标签信息的协同过滤算法做出进一步地改进。对用户的评分和标注行为加入相应的时间权重因子,以此降低时间间隔较长的评分和标注行为在相似度计算时的影响权重,从而使得推荐结果能更加符合用户的兴趣变化。
研究结果 —— 首先,对社会化标签进行预处理,根据使用的频繁性选出真正有区分价值的社会化标签用于后续计算。并通过两种标签信息的调节因子取值进行比对实验来确定其合适取值,使得基于标签相似度的计算结果更加准确;其次,将标签信息和协同过滤推荐算法进行融合。通过与传统协同过滤和基于标签推荐的算法比对实验,来验证融合标签信息的协同过滤改进算法的可行性;最后,引入时间信息进行不同遗忘速度取值比对实验来确定其合适取值,并通过与融合标签信息的协同过滤算法的比对实验来验证基于时间信息改进算法的可行性。
研究的局限性 —— 仅根据使用的频繁性来选择具有区分价值的社会化标签。但社会化标签是用户凭借自由意志使用,且没有规范的词汇和层次结构的限制。因此可能存在一词多义和同义词等现象,导致社会化标签中隐含的信息不能很好地被理解。此外,该算法在分类的效率和准确度上也有待进一步提升。
实际影响 —— 通过融合标签信息的协同过滤算法,使用标签来加强用户和资源之间的联系,提高推荐的准确度。并通过引入时间权重因子来把握用户兴趣的转移,为用户提供更加有效地推荐,提高用户与系统的黏着性。
独创性 —— 将标签划分为标准化标签和社会化标签,针对两种标签所反映的不同信息提出基于标签信息相似度计算的改进公式;融合标签信息和时间信息对传统协同过滤方法进行算法改进,使得推荐结果更加符合实际情况,并能在一定程度缓解数据稀疏、用户兴趣转移等问题。
Other AbstractPurpose — Recommendation system is an effective way to solve the problem of information overload, which can help users quickly find resources that may be interest. However, there are still some problems in the traditional recommendation system. Such as, data sparseness of scoring matrix, transfer of user interest preferences and cold start. Tag information has both resource description information and user interest preference information, its related applications are more and more widely used. It can be used not only for the recommendation of conventional resources such as movies and books, but also for the recommendation of social networking. It brings a new turn for traditional recommendation systems. However, the use of tag information also brings new problems, such as how to extract valid information from tags. For this reason, tag and time information are integrated into the traditional collaborative filtering algorithm. Strengthen the relationship between users and resources in the form of tags, strengthen the relationship between recommendation and user interest preferences in the form of time.
Design/methodology/approach — Firstly, according to the information of resource description and user interest preference reflected by tags, tags are classified into standardized tags and social tags. The formula for calculating tag similarity is improved. Secondly, tag information is introduced into the basis of traditional collaborative filtering. The improved tag similarity calculation formula is used to predict the first score value, and the score prediction value is filled in the original score matrix. The user's nearest neighbor recommendation is made on the filled matrix to get the final recommendation result. Finally, aiming at the user interest transfer problem, the collaborative filtering algorithm which fuses tag information is further improved. The time weighting factor is added to the user's rating and tagging behavior to reduce the influence weight of the longer time interval rating and tagging behavior in the similarity calculation, so that the recommendation results can be more in line with the user's interest changes.
Findings — Firstly, the socialized tags are preprocessed. According to the frequency of usage, the socialized tags with real distinguishing value are selected for subsequent calculation. The appropriate values of the two kinds of tags are determined by comparing the values of adjustment factors, which makes the results based on tag similarity more accurate. Secondly, tag and collaborative filtering recommendation algorithm are fused. The feasibility of the improved collaborative filtering algorithm based on tag information is verified by comparing with traditional collaborative filtering algorithm and tag-based recommendation algorithm. Finally, introducing time information to compare the values of different forgetting speeds to determine the appropriate values, and the feasibility of the improved algorithm based on time is verified by comparing with the tag-based collaborative filtering algorithm.
Research limitations/implications — Social tags with differentiating value are selected only according to the frequency of use. However, in view of the fact that social tags are used by users with their free will and without the restriction of standard vocabulary and hierarchical structure, there may be polysemy and synonyms, which may lead to the incomprehension of the implicit information in social tags. In addition, the efficiency and accuracy of the algorithm need to be further improved.
Practical implications — Through the collaborative filtering algorithm which fuses tag information, use tags to strengthen the links between users and resources, improve the accuracy of recommendation. By introducing time weighting factor to grasp the transfer of user’s interest, provide more effective recommendation for users, enhance the cohesion between users and the system.
Originality/value — The tags are divided into standardized tags and social tags, and an improved formula based on similarity calculation of tag information is proposed according to the different information reflected by the two tags. The traditional collaborative filtering method is improved by integrating tag information and time information, which makes the recommendation results more realistic and can alleviate the problems of data sparsity and user interest transfer to a certain extent.
Document Type学位论文
First Author AffilicationSchool of Information Science and Engineering
Recommended Citation
GB/T 7714
李雅琴. 融合标签和时间信息的协同过滤推荐[D]. 兰州. 兰州大学,2019.
Files in This Item:
There are no files associated with this item.
Related Services
Recommend this item
Usage statistics
Export to Endnote
Altmetrics Score
Google Scholar
Similar articles in Google Scholar
[李雅琴]'s Articles
Baidu academic
Similar articles in Baidu academic
[李雅琴]'s Articles
Bing Scholar
Similar articles in Bing Scholar
[李雅琴]'s Articles
Terms of Use
No data!
Social Bookmark/Share
No comment.
Items in the repository are protected by copyright, with all rights reserved, unless otherwise indicated.