Result Disambiguation in Web People Search

Abstract

We study the problem of disambiguating the results of a web people search engine: given a query consisting of a person name plus the result pages for this query, find correct referents for all mentions by clustering the pages according to the different people sharing the name. While the problem has been studied extensively, we discover that the increasing availability of results retrieved from social media platforms causes state-of-the-art methods to break down. We analyze the problem and propose a dual strategy where we distinguish between results obtained from social media platforms and those obtained from other sources. In our dual strategy, the two types of documents are disambiguated separately, using different strategies, and their results are then merged. We study several instantiations for the different stages in our proposed strategy and manage to achieve state-of-the-art performance.

Person Name Disambiguation in the Web Using Adaptive Threshold Clustering

“…The most popular types of features were: bag of words, named entities (NEs), and noun phrases. These features were usually weighted using the well‐known term frequency‐inverse document frequency (TF‐IDF) scheme (Berendsen et al, ; Chen & Martin, ; Liu, Lu, & Xu, ; Yoshida, Ikeda, Ono, Sato, & Nakagawa, ). Balog et al () to compare a VSM representation with respect to using probabilistic Latent Semantic Indexing (pLSI), a topic model representation, showing that the first option reaches significantly better results.…”

“…Regarding the corpora for evaluating PND in a web search scenario, in addition to the WePS datasets that we use in the experiment and are described in the Results and Discussion section, we can mention the corpus presented by Berendsen et al (), which was used to study the impact of social network webpages in this problem. However, the rankings contained in this corpus were not made up of the results of a query to a search engine, but the authors selected search results from several search engines to include mostly webpages from social platforms, which leads to a corpus that does not reflect a real query output.…”

“…For instance, some webpages related to a certain individual could be professional sites (e.g., corporation webpages), while others may contain personal information (e.g., blogs, social profiles, and so on) and both kinds of webpages could share little common vocabulary. Berendsen, Kovachev, Nastau, de Rijke, and Weerkamp () conclude that the appearance of social networking platforms increases the difficulty of this problem.…”

Bootstrapping Wikipedia to answer ambiguous person name queries

“…For each of the methods, we tested many different configurations. For example, for fairness reasons, one of the configurations of K-Means used the Wikipedia entities as initial centroids similarly as proposed in [8]. However, random seeds 7 led to better results than the above 6 http://www.cs.waikato.ac.nz/ml/weka/ 7 We used the average results over ten repetitions.…”

Text Document Topical Recursive Clustering and Automatic Labeling of a Hierarchy of Document Clusters

“…This representation allows users to effectively zoom in and locate the documents of interest. It has been proved to facilitate the searching and browsing process [1]. This paper borrows Community Mining from Social Network Analysis to discover different topical coherent document groups and gives each document cluster descriptive labels.…”