The c-regression model is a method that simultaneously performs clustering and regression to obtain regression equations for each cluster and express the overall structure of the dataset. Gaussian Process c-Regression Models has been proposed as a method to extend the c-regression model to nonlinear models. In this paper, we propose Gaussian Process Sequential Regression Models, which do not require the number of clusters and can obtain nonlinear regression models. The optimization of kernel parameters used in the proposed method using the gradient method, or MCMC was also studied. The experimental results suggested that the proposed method outperforms the conventional clustering methods in terms of the maximum and average values of ARI.

With a view to resolving mismatches between the questioners and respondents at Question and Answer (Q&A) sites, factor scores were estimated using feature values of statements, on the basis of the scores of nine factors obtained experimentally. A method of selecting respondents who can appropriately answer a question was then proposed. As a result of analysis, the proposed method successfully selected potential respondents that were more than approximately average when it came to appropriately answering a given question. Nevertheless, this methodology was greatly dependent on the syntactic information extracted through morphological analysis. Therefore, N-gram has been applied to the methodology as an alternative method of morphological analysis. In applying N-gram, the estimation of factor scores and detecting of the respondent who can appropriately answer a question have been realized so far. Thus, in this paper, finding respondents who can properly answer a newly posted question based on 2-gram is investigated. An experiment is conducted to evaluate three sets of 100 answer statements extracted in accordance with the Euclidean distances based on 2-gram. As a result of the analysis, the proposed method utilizing 2-gram can choose potential respondents that were more than approximately average when it came to answering appropriately. Therefore, it has also been shown that the proposed method using 2-gram would also be applicable.

In order to solve the issues of mismatches between the intentions of questioners and respondents at Question and Answer (Q&A) sites, nine factors of impressions for Q&A statements were obtained through factor analysis applied to the results of impression evaluation experiments. Then through multiple regression analysis, factor scores were estimated by using the feature values of statements. The factor scores estimated and obtained were subsequently utilized for detecting respondents who are expected to appropriately answer a posted question. Nevertheless, up to now the meanings and contents of Q&A statements have not been taken into consideration. Therefore, this paper aims to consider the semantics between Q&A statements. The feature values are reviewed and narrowed down to syntactic information, closing sentence expressions, 2-gram, and word2vec. The analysis result conveys that all the trials show good estimation with the consideration of cross-validation. It has also been suggested that applying word2vec could play a vital role in estimating improved factor scores.

In order to detect the capability deterioration of economic activity for single elderly people at age of sixty-five or over, characteristic expenditures are tried to be extracted. The data used are anonymous and obtained from the National Survey of Family Income and Expenditure carried out by Ministry of Internal Affairs and Communications in 2004. As a basis of economy activity, clustering is performed with five clusters having income and saving as feature values. We then develop a method to determine whether savings are enough for life expectancy in terms of income and saving. As a result of clustering, data (13 males, 48 females) determined as having dissaving risk are included in the corresponding clusters (32 males, 226 females) categorized as low or fairly low income and low saving. In the case distinguishing rent with own house and using four clusters having saving as feature value, Correct Detection Ratio (CDR) of elderly people determined as having dissaving risk is 66.7% (83.3%, 94.1% and 52.6%, respectively) for Male/Rent (Male/Own, Female/Rent and Female/Own). In the case disregarding rent and own house, CDR is 78.6% (65.5%, respectively) for male (female). For female, further consideration of clustering conditions results in improving CDR as high as 73.9%.

In order to detect the capability deterioration of economic activity for elderly people at age of sixty-five or over, we have used an anonymous data obtained from the National Survey of Family Income and Expenditure (NSFIE) carried out by Ministry of Internal Affairs and Communications (MIAC). We develop a method to detect dissaving risk of elderly people. So far the analysis data were divided into test data and training data. Then three kinds of methods were performed in terms of income and savings. Two-step methods were taken to determine dissaving risk. In using anonymous data, however, there is controversy if anonymity is secured. In order to strengthen the anonymity of the data, in this paper, random data is generated with using the anonymous data and then compared with the case of analyzing anonymous data as it is, with a view to performance evaluation. As a result of analysis, it could be concluded that using the random data would be as effective as using anonymous data for evaluating the performance of the proposed method.

To eliminate mismatches between the intentions of questioners and respondents of Question and Answer (Q&A) sites, we have clarified that the impression of the statements could be captured by nine factors, and the factor scores could be estimated from the feature values of the statements. Objective scores of the statements could be estimated fairly good, and that those of subjective statements could be estimated well by taking the natural logarithm of the factor scores with the consideration of cross-validation. This paper tries to perform multiple regression analysis with the consideration of semantic similarity between Q&A. A statement is represented with an average word vector of the words appearing in the statement. The semantic similarity of two statements is calculated through the cosine of the two average word vectors of the two statements. As a result, there could be a possibility that semantic similarity would be effective in estimating objective scores.

In order to eliminate mismatches between the intentions of questioners and respondents of Question and Answer (Q&A) sites, nine factors of impressions for Japanese statements have experimentally been obtained. Nine factors have also been obtained from the impression of English Q&A statements. This paper estimates factor scores of English Q&A statements through multiple regression analysis. These are words and characters, syntactic information, and appearance percentages. It is shown that estimation accuracies of all of the nine factors are very good.

This paper proposes a method for selecting respondents who can give an appropriate answer to a question, in order to eliminate mismatches between the questioners and respondents at question-and-answer sites. The possibility of detecting respondents capable of appropriately answering a newly posted question is examined and confirmed. Based on this, the proposed method uses the number of appearances of each respondent and scores them based on the distance between the factor scores of a question and previously posted answers. Impressions of statements were obtained for nine factors. Factor scores were estimated using multiple regression on the feature values of the statements. This method was evaluated by experiments that compared its precision and recall with those of methods based on average scores and distances. It was shown that the proposed method outperforms these other methods. It is also shown that, for a given question, the proposed method can successfully select those respondents who are more appropriate than around average.

To eliminate mismatches between the intentions of questioners and respondents of Question and Answer (Q&A) sites, we have clarified that the impression of the statements could be captured by nine factors, and the factor scores could be estimated from the feature values of the statements. Objective scores of the statements could be estimated fairly good, and that those of subjective statements could be estimated well by taking the natural logarithm of the factor scores. This result, however, is likely to estimate inaccurate values due to overfitting to the explanatory variables. This could result from not having considering cross-validation through the attempts. This paper tries to perform multiple regression analysis with the consideration of cross-validation. For statements of Yahoo! Auction, PC and love counseling & human relationships, respective five trials have been attempted. As a result, most of the attempts show good or fairly good estimation accuracy. It is suggested that estimation errors should be reduced.

To eliminate mismatches between the intentions of questioners and respondents of Question and Answer (Q&A) sites, we have clarified the characteristics of the question and answer statements. It has been shown that the impression of the statements could be captured by nine factors, and the factor scores could be estimated from the feature values of the statements. Here, the objective scores of answer statements are provided. This paper tries to estimate the objective scores of answer statements through multiple regression analysis. They are estimated from the factor scores estimated by using multiple regression formulas already obtained. Absolute values of the differences between the factor scores of answer statements and those of question ones, those of the answer statements having the highest scores, and those of the ones having the lowest ones, are used as the explanatory variables. It is shown that objective scores of the objective statements could be estimated fairly good, and that those of subjective statements could be estimated well by taking the natural logarithm of the factor scores.

In order to avoid the problem of mismatch between the questioner and the respondent, we have conducted impression evaluation experiment and nine factors are obtained as a result. Then factor scores of any other statements have been tried to be estimated by using multiple regression analysis@from feature values of statements. By adopting syntactic information of statements, word image ability and closing sentence expressions as the feature values, all the factor scores were well estimated. This paper tries to explain the estimation result with major feature values. The explanation is confirmed by comparing the features of the statements having high and low factor scores with the major features.