【作者】 李林；

【导师】 姚兴苗；

【作者基本信息】 电子科技大学 ， 通信与信息系统， 2014， 硕士

【摘要】 文本推荐是指当用户确定兴趣模型后,系统能对比、判断目标文本与用户兴趣模型的相似程度,并按照相似度由大到小的次序向用户推荐所需文本。常用的文本推荐方法是基于文本向量空间中的特征词字面形式上的严格匹配。在目标文本的特征词集合与用户兴趣模型的特征词集合中,相同特征词的个数越多,目标文本与用户兴趣模型就越相似,它被系统推荐的优先级就越高;相同特征词的个数越少,目标文本与用户兴趣模型就越不相似,它被系统推荐的优先级就越低。但由于自然语言的灵活性,不同的文本对同一事物的描述可能会使用意义相同的同义词。如果在文本推荐的过程中,仅仅考虑特征词字面形式上的匹配,而不考虑同义词匹配,必然出现推荐结果单一、覆盖不全面、准确率不高的问题,从而无法实现向用户准确推荐其所需文本的目的。本文首先对目前的文本推荐方法进行了学习和分析,重点研究了数值型向量空间模型。其次,本文提出了一种基于同义词的文本推荐的改进方法。该方法既考虑特征词字面匹配,同时又考虑同义词语义匹配对文本推荐的积极作用。这样,文本推荐过程就不再只是简单的字面匹配,而是一种语义的匹配。这可以在一定程度上提高文本推荐的精度。通过对实验结果的分析可知,这种改进方法可以更加准确地向用户推荐所需的文本,正确率最高可提高40%。最后,将链路预测方法运用到同义词网络中,利用复杂网络的思想,将同义词构建成网络,然后进行链路预测,预测出目前同义词网络中没有的但确实含有同义词关系的新的同义词对。然后将这部分新的同义词关系与已有同义词关系进行合并,使网络中单词之间的同义词关系更加丰富。新预测出的同义词对使得整个同义词网络的覆盖范围更加广泛。在文本推荐过程中,用户兴趣模型与目标文本同时出现新的同义词关系时,继续使用合并后的同义词网络,可以进一步提高文本推荐的性能。更多还原

【Abstract】 Text recommendation system means that when a user interest model is determined, the system can be compared to determine the similarity of the target text and user interest model and recommend the necessary text to the user according to the similarity in descending order.The recommended method is commonly used text matching strictly text-based feature vector space in the form of the word literally. The more number of the same features of the word set between target text and topic text, the target text is more similar with the theme text, and it is recommended by the system more higher priority; The fewer number of the same features of the word set between target text and topic text, the target text is less similar with the theme text, and it is recommended by the system more lower priority. However, due to the flexibility of natural language, different text description of the same thing will use different synonyms. Without considering the synonym match, the inevitable recommended the results of the single coverage, not comprehensive, accuracy is not high, thus the system is unable to achieve the goal of recommending text of user required accurately, if in the process of text recommended only consider the key match in the form of the literal.Firstly, the article has carried on the study and analysis the current text recommended method, focusing on the numerical vector space model.Secondly, we propose a synonyms-based text recommended improvements. The method not only considered key words literally match, while considering the positive role of synonyms semantic matching in process of text recommendation. Thus, the process of text recommendation is no longer just a simple literal match, but a kind of semantic matching. To some extent, it can improve the accuracy of text recommendation. The experimental results show that the method can recommend the desired text to the user more accurately, the accuracy increases of most 40%.Finally, this paper proposes a synonym network-based link prediction method. Using the ideas of complex network, the method will build a synonym network, then predict the link, which not in the current network but does contain a new synonym relation. We can merge this part of the new synonym relationship and existing synonyms. The synonymous relationship between words in new network will be more rich. Then this part of the new link of synonym relationship, make the semantic matching coverage of some individual words more widely. Continuing to use the combined synonym network, can further improve the performance of text recommendation, when new synonym relation appear the user interest model and the target text at the same time.更多还原