Combining Mention Context and Hyperlinks from Wikipedia for Named Entity Disambiguation

Abstract

Named entity disambiguation is the task of linking entity mentions to their intended referent, as represented in a Knowledge Base, usually derived from Wikipedia. In this paper, we combine local mention context and global hyperlink structure from Wikipedia in a probabilistic framework. Our results show that the two models of context, namely, words in the context and hyperlink pathways to other entities in the context, are complementary. We test our method in eight datasets, improving the state-of-the-art results in five, without any tuning, showing that it is robust to out-ofdomain scenarios. When tuning combination weights, we match the best reported results on the widely-used AIDA-CoNLL test-b.

Combining Textual and Graph-Based Features for Named Entity Disambiguation Using Undirected Probabilistic Graphical Models

“…By combining different sources of information comprising knowledge about entities, names, context, and the Wikipedia graph in a probabilistic framework, Barrena et al [3] observe complementary effects between these features. However, they impose strong independence assumptions (i) on the level of features, which essentially renders their model an instance of Naïve Bayes classification, and (ii) on the level of entities as well.…”

“…(2) and obtain a probability distribution over all generated states. 3 We select a single candidate state s t by sampling from the distribution of generated states 4 to obtain a potential successor state. We accept the sampled successor state s t as our next state s t+1 if it has a higher probability than the previous state s t :…”

With a Little Help from my Neighbors

“…The framework in [11] is similar to our methods as it determines characteristic keyphrases for each person entity and compares them to the context of a person mention. Another comparable approach [2] makes use of hyperlinks from Wikipedia and mention context using a dictionary, a graph, and textual context. The dictionary maps surface forms to a Wikipedia articles.…”

“…This is similar to our approach but instead of using anchor texts, which need to be extracted from different Wikipedia pages, we make only use of Wikidata labels and alternative names. The graph in [2] is built from the Wikipedia link structure. In our approach, however, we calculate similarities between two person in the network by the number and distance of their wikilink co-occurrences within Wikipedia articles.…”

“…In our approach, however, we calculate similarities between two person in the network by the number and distance of their wikilink co-occurrences within Wikipedia articles. Personalized PageRank and random walks are used in [2] to assign a Wikipedia article to a person mention. Our approach, in contrast, only relies on a social network that is constructed from co-occurrences of persons, leading to more efficient neighborhood computations and in particular better disambiguation results.…”