基于矢量空间模型和最大熵模型的词义问题解决策略

针对单义词的词义问题构建了融合触发对(trigger pair)的矢量空间模型用来进行词义相似度的计算，并以此为基础进行了词语的聚类；针对多义词的词义问题应用融合远距离上下文信息的最大熵模型进行了有导词义消歧的研究。为克服以往词义消歧评测中通过人工构造带有词义标记的测试例句而带来的覆盖程度小、主观影响大等问题，将模型的评测直接放到了词语聚类和分词歧义这两个实际的应用中。分词歧义的消解正确率达到了92％，词语聚类的结果满足进一步应用的需要。

Solution Strategies for Word Sense Problems Based on Vector Space Model and Maximum Entropy Model

Aiming at the monosemantic problem, the vector space model with trigger pair was built to calculate the semantic similarity. Based on this, the word clustering was carried out. For polysemantic words, the maximum entropy model with long distance information was applied to word sense disambiguation. Because manually tagging the test sentences with semantic tag for word sense disambiguation evaluation will bring the problem of small size and uncertainty, the evaluation was put into word clustering and Chinese word segmentation system directly. The precision of Chinese word segmentation disambiguation is 92% and the word clustering can satisfy the further requirements.