Institutional Repository
| 基于贝叶斯推理的点击模型及其实现 | |
| Alternative Title | click model based on bayesian inference and its implementation |
| 孙付伟; 李娟; 杨达 | |
| 2013 | |
| Source | 计算机应用与软件
 |
| ISSN | 1000-386X |
| Volume | 30Issue:1Pages:7-10 |
| English Abstract | 为能更好地解释搜索引擎和商务搜索的点击日志中的用户行为,实现一种用于分析日志中包含的用户行为的贝叶斯点击模型。通过分析中国最大电子商务网站的约927万条用户搜索点击日志数据,发现一个的文档的点击是受其上下位置点击过的文档共同影响的,然后基于此发现提出并实现一种新的基于贝叶斯推理的点击模型,并给出并行版本的算法实现。最后通过利用来自用户搜索的一个月日志数据验证,结果表明该模型优于现有的点击模型。 |
| Indexed Type | CNKI ; WANFANG ; CSCD |
| Abstract | In order to better explain user behaviour from click logs in search engine or sponsored search, we implement a Bayesian click model for analysing user behaviours included in logs. By analysing about 9.27 million click log data collected from a largest e-commerce site of China, there finds that the click probability of a document is affected by the clicked documents above and below it. Then we propose and implement a new click model based on Bayesian inference according to the phenomenon found, together with the implementation of an algorithm in parallel version. At last, we validate the model through a log data set collected about a mouth from user search, and the result shows that the proposed model outperforms existing click models. |
| Keyword | 点击日志 点击模型 贝叶斯推理 搜索引擎 日志分析 |
| Department | 中国科学院软件研究所基础软件国家研究中心;中国科学院研究生院; |
| Subject | Computer Science (Provided By Thomson Reuters) |
| Sponsorship | 国家自然科学基金项目(60903050,61100071)|国家重点基础研究发展计划基金(2007CB310802)|国家重大科技专项经费资助项目(2010ZX01036-001-002,2010ZX01037-001-002) |
| Language | 中文 |
| CSCD ID | CSCD:4737451 |
| Content Type | 期刊论文 |
| URI | http://ir.iscas.ac.cn/handle/311060/15238 |
| Collection | 中国科学院软件研究所 |
| Recommended Citation GB/T 7714 | 孙付伟,李娟,杨达. 基于贝叶斯推理的点击模型及其实现[J]. 计算机应用与软件,2013,30(1):7-10. |
| APA | 孙付伟,李娟,&杨达.(2013).基于贝叶斯推理的点击模型及其实现.计算机应用与软件,30(1),7-10. |
| MLA | 孙付伟,et al."基于贝叶斯推理的点击模型及其实现".计算机应用与软件 30.1(2013):7-10. |
| Files in This Item: | There are no files associated with this item. | |||||
Items in the repository are protected by copyright, with all rights reserved, unless otherwise indicated.
Edit Comment