Highly stable term co-occurrence model

To address the issues that traditional term co-occurrence models are lack of theoretical basis and poor stabile, a highly stable term co-occurrence model based on term field is proposed. The model uses the concept of field in classical physics for reference to define the term field (terms are the basic units of language, which describe the abstract concepts, and the term field is the area affected by a term in document). Based on the definition, the model regards correlation as a superposition of term fields, and gets the functional relations of terms correlation and the distances between terms. Experimental results show that the terms correlation in this model is almost a constant for small distances and stable enough in window. While the correlation amplitude of the terms in same category is only 26% of the best result obtained with other models, which means the model is stable enough in dataset.