Lin and Wu. 2009. Phrase Clustering for Discriminative Learning.
Phrase clustering for discriminative learning, by D. Lin, X. Wu. In Proceedings of the Annual Meeting of the Association for Computational Linguistics and the International Joint Conference on Natural Language Processing of the Asian Federation of Natural Language Processing, 2009.
This paper makes use of phrase clustering to improve on the state of the art for the Named Entity Recognition problem. They obtained 1.0 F-score improvement over NER systems on the CoNLL benchmark (in 2009). In their paper, they used phrases that occur more than 100 times in a 700 billion token web corpus (Lin et al., 2008).
Brief description of the method
Due to the large number of possible phrases, the authors used Bloom filters to decide whether a sequence of tokens is considered a phrase.
Phrases as feature vectors
Each phrase is represented as a vector of its context words. The frequency count of words appearing within a fixed-sized window is aggregated and converted into pointwise mutual information(PMI) values.
Parallel K-Means using MapReduce
The phrase vectors are then clustered using K-means clustering algorithm. The authors chose the K-means algorithm over other more advanced clustering algorithms because it is fast and can be very easily parallelized. Given the amount of data they have, this is a very important consideration. It took about 20 minutes on a 1000 machine cluster to perform K-means to convergence on their dataset. The "similarity" between two phrases is simply the Euclidean norm of the phrase vectors in the feature space. What this means is that phrases are considered to be more similar if they share more context words.
In addition to just performing hard clustering (where each phrase can only belong to one cluster), they perform soft clustering as well by assigning phrases to cluster centroids that are within a threshold distance. The motivation behind doing soft clustering is that it may be better able to model the fact that phrases may contain several "senses".
The effectiveness of phrase clustering is evaluated on Named Entity Recognition. For NER, they used a 1-word context window and hard clustering, and a linear chain CRF with standard features used in current NER systems. The baseline features contains a total of 48 feature templates.
The results on CoNLL'03 test set are as follows:
The authors evaluated their phrase clustering system on the KDDCup 2005 query classification competition. The task is to categorize 800k internet user search queries into 67 topical categories. They treated the problem as 67 separate binary classification task and trained logistic regression classifiers with regularization. Their system were on par with the winning KDDCup 2005 system.
Brown et al, CL 1992 introduced the Brown clustering algorithm which is frequently used in NER for clustering words. It is based on the idea that one should minimize the bigram language model perplexity of a text corpus.
Miller et al, ACL 2004 used a word clustering algorithm and discriminative training approach to the NER problem.