横山 友也

This paper tries to find the possibility of searching respondents capable of giving appropriate answers to a newly posted question in order to eliminate mismatches between the intentions of questioners and respondents of Question and Answer sites. The impressions of sixty questions and their responses posted at Yahoo! Chiebukuro have been experimentally evaluated. By applying factor analysis to the scores of the experiment, the impression of the statements could be captured by nine factors. Factor scores can be estimated by using the features including syntactic information, word imageability, closing sentence expressions, word familiarity, and notation validity. The good overall estimation accuracy has been obtained. The distances between the factor score of a question statement and those of answer statements already posted are calculated. It is shown that the distance and the number of appearance may help us select users who can give appropriate answers to a question. With the consideration of categories of answer statements, users who specialize in certain categories could be regarded to give more appropriate answers.

This paper proposes a method of selecting respondents who can appropriately answer a question to eliminate mismatches between the questioners and respondents at Question and Answer sites. The proposed method uses the number of appearance of respondents and the score based on the distance between the factor scores of a question and an answer already posted. Nine factors of impressions for statements have experimentally been obtained. Factor scores have been estimated through multiple regression. The possibility of detecting respondents capable of appropriately answering a newly posted question has been examined. The proposed method is based on this observation. It is experimentally evaluated by comparing it with the methods based on average scores and distances through precision and recall. It is shown that the proposed method outperforms others. It is also shown that the proposed method could successfully select the respondents that are more than almost averagely appropriate to a question.

In order to avoid the problem of mismatch between the questioner and the respondent at Q & A sites, we have experimentally evaluated the impression of 60 question and answer statements posted at Yahoo! Chiebukuro by using 50 impression words. Nine factors as to statements have been obtained by applying the factor analysis to the scores obtained through the experiment. Factor scores of any other statements have been tried to be estimated by using multiple regression analysis. This paper tries to estimate those factor scores through the multiple regression analysis by adopting such feature values as syntactic information, word imageability and closing sentence expressions. As a result, estimation accuracy for every factor is overall well and drastically improved by focusing on closing sentence expressions as well as syntactic information and word imageability. The validity of estimation is also confirmed by inspecting important feature values.