This thesis proposes a word recommendation system using information retrieval techniques. The concept of the proposed system is the opposite of the dictionary system which looks up a word with its meaning. That is, the input of the proposed system is the definition or the meaning of a word and the output of the system is the ranked words which are most proper for the meaning (input: meaning, output: word).
In the proposed system, documents are definitions or meanings of words and document identifiers are words themselves, compared to information retrieval systems. In this case, the length of the documents is too short because the definitions in the dictionary is too short and then the performance of the proposed system might be degraded. To alleviate the problem, in this thesis, methods for document expansion: glossary expansion, synonym expansion, thesaurus expansion are used. The proposed system computes the similarity between expanded documents and user queries (the meanings of words) and recommends words using the several methods for document expansion.
The performance (r-inclusion rate) amounts to almost 100% when the queries are meanings of words in the dictionary, and to 72.1% when the queries are meanings which users write in person. Through the several experiments, we have observed that the document expansion is very useful for the word recommendation system.
In the future, new measures including the r-inclusion rate of our proposed measure are required for performance evaluation of word recommendation systems and new evaluation sets for objective assessment. Furthermore, antonyms as well as synonyms are needed for document expansion.