2018年11月28日讨论班报告

各位老师，同学：
时间：11月28日讨论班 15:30-17:00
地点：教学楼5-308
报告内容：

邵礼旭Toward an Interactive Patent Retrieval Framework based on Distributed Representations

专利检索是专利分析任务的关键，本文提出了一种简单实用的交互式关联反馈机制，让用户在前n次点击中注释相关/不相关的检索结果。然后，利用这一反馈来进行查询重构和词语加权，根据每个词在区分相关候选专利和无关候选专利方面的能力来分配权重。本文演示了分布式表示在CLEF-IP 2010数据集上的有效性，比关键字搜索基线的召回率显著提高了4.8%。其次，本文模拟用户交互来演示交互式词语加权机制的有效性。仿真结果表明，在进行一次交互迭代后，本文的方法可以比传统的语义检索和交互式专利检索方法在召回方面取得显著改进。

杨哲A Structural Intention Recognition and Matching Method for User-Generated Short Texts

在互联环境下，社交网络中存在信息过载的问题，用户的供需意图匹配存在挑战，文章设计了一种结构化的意图识别与匹配的方法。将一个意图表示为意图词，意图对象，意图情感和约束四个部分，针对每个部分给出了识别与匹配算法，并且结合各部分优先级和权重给出了匹配度算法。另外，还研究了在海量数据中快速进行供需意图的匹配的方法。

    Online social media platforms are becoming increasingly popular. A huge number of people are using such platforms to share various information. Supply and demand constitute two important kinds of information in social media platforms. Examples are sell and buy, positions and job seeking, houses for rent and needs of accommodation, and so on. Consequently, it is relevant to match such supply and demand information automatically to facilitate the trades online. However, the problem is challenging due to the high volumes of information expressed in short and vague texts. In this paper, we propose a complete set of techniques to resolve the problem. First, we design an intention recognition model to recognize a user’s complete intention from her/his original text posts. The recognized intention includes expected behaviors, intention objects, user sentiments and other relevant constraints. Second, we define the similarity for two intentions that are recognized. Third, we devise an algorithm to match supply and demand intentions whose similarity is sufficiently high. We evaluate our proposal on real data. The results show that our approach returns accurate matching results efficiently.

下周报告人 殷小静，朱方林