| Summary: | 碩士 === 元智大學 === 資訊工程學系 === 106 === Analogical reasoning is a reasoning method which based on a certain kind of thing has some kind of attribute, to speculate that similar things have the same attribute. Assuming that there are four objects: a, b, a*, b*, analogical reasoning means that the relationship between a and b is consistent with the relationship between a* and b*, expressed as "a: b :: a*: b*". For example, there are two pairs of words ("China", "Beijing") and ("Japan", "Tokyo"), there exists the analogical relationship between them, which is the state and the capital, constituting a language pattern. The combination of analogical reasoning and the popular deep learning word embedding models, has produced surprising results in the field of natural language processing.
With the expansion of the number of people suffering from depression, triggering a high degree of attention from all walks of life. Negative life events are an important reason of causing depression, such as the death of family members, quarrel with the spouse, fired by the boss or blamed by the teacher. The combination of a subject and a negative life event is called as a language pattern of negative life events. Therefore, whether it can through identify these negative life event language patterns automatically and accurately to understand those web text with depression trend, which is important to establish effective and practical psychiatric network services.
In this study, we focus on negative life event language patterns, applying three different language models which are Hyperspace Analog to Language model, Skip-gram model and the Continuous Bag-of-Words model respectively, to do the survey of language patterns analogical reasoning. The study can be divided into three parts: 1) training word embedding models with Wikipedia Chinese corpus; 2) finding out the existence of negative life event language patterns from the depression Q&A texts, such as ("parents", "divorce"), and then generate query data set; 3) combining different word representations and different methods of analogical reasoning to do experiments of negative life event language patterns mining with analogical reasoning.
|
|---|
