Assigning Belief Scores to Names in Queries

Assigning Belief Scores to Names in Queries Christopher Dozier Research and Development Thomson Legal and Regulatory 610 Opperman Drive Eagan, MN 55123,USA chris.dozier@westgroup.com ABSTRACT Assuming that the goal of a person name query is to find references to a particular person, we argue that one can derive better relevance scores using probabilities derived from a language model of personal names than one can using corpus based occurrence frequencies such as inverse document frequency (idf). We present here a method of calculating person name match probability using a language model derived from a directory of legal professionals. We compare how well name match probability and idf predict search precision of word proximity queries derived from names of legal professionals and major league baseball players. Our results show that name match probability is a better predictor of relevance than idf. We also indicate how rare names with high match probability can be used as virtual tags within a corpus to identify effective collocation features for person names within a professional class. 1. INTRODUCTION Some of the most common types of queries submitted to search engines both on the internet and on proprietary text search systems consist simply of a person’s name. To improve the way such queries are handled, it would be useful if search engines could estimate the likelihood or belief that a name contained in a document pertains to the name in the query. Traditionally, relevance likelihood for name phrases has been based on inverse document frequency or idf, [3][4]. The idea behind this relevance estimate is that names which rarely occur in the corpus are thought to be more indicative of relevance than names that commonly occur. Assuming that the goal of a person name query is to find references to a particular person, we argue that one can derive better relevance scores using probabilities derived from a language model of personal names than one can using corpus based occurrence frequencies. The reason for this is that finding references to a particular person in text is more dependent upon the relative rarity of the name with respect to the human population than it is on the rarity of the name within a corpus. To get an intuitive idea of this point, consider that, within a corpus of 27,000 Wall Street Journal articles published between January and August of the year 2000, the name “Trent Lott” occurred in 80 documents while the name “John Smith” occurred in 24. All 80 references to “Trent Lott” referred to the majority leader of the U.S. Senate, while “John Smith” references mapped to 5 different people. This is not surprising. From our experience, we know that “Trent Lott” is an uncommon name and “John Smith” is a common one. We present here evidence that name match probability based on a language model predicts relevance for name queries far better than idf. It may be argued that idf was never intended to be used to measure the relative ambiguity of a name query. However, idf …

Content truncated to 500 words