This paper describes an MUSAI (MUSical expression generation system by Adjective with Interaction) reflecting user's impression represented by an adjective, i.e., an image word. The MUSAI uses a Kansei space and fuzzy rules that represent a relationship between musical expression and impression. The inputted user's impression is mapped into a Kansei space and the parameter values of musical expression are obtained by fuzzy inference. Then, the musical expression is generated based on the obtained parameter values and the inputted MusicXML. If a user is not satisfied with generated musical expression, a user inputs a modification word with an adverb and an image word such as more bright, and the musical expression is modified by changing the coordinates values in a Kansei space. The validity of the MUSAI is confirmed by experiments. From the experimental results, it is found that the MUSAI successfully generates musical expressions reflecting each experiment participant's impression.
New products are frequently launched in the world. Most of new products could be regarded as combinations of characteristics of existing products. All the combinations do not always become seeds of new selling products. Each existing product has functions. A new function may be thought up by combining two functions of existing products. If a new function is novel, a combination may be a seed of a new selling product. This paper proposes an evaluation method of novelty of new functions obtained from product functions for supporting idea generation. The system makes phrases representing new functions, and evaluates whether each of new functions is novel or not using Web hit counts. The system outputs products combinations with novel functions. We conducted evaluation experiments, and verified that the proposed system can support participants in idea generation.
In the field of marketing, it is very important to design a marketing strategy based on the acquired data from customers. Statistical analysis techniques called data mining are utilized to acquire useful information from large amounts of data. Recently, customer groups with features different from others called minority groups are emphasized because of Customer Relationship Management (CRM). Thus, it is strongly needed to extract and analyze minority groups from customer data such as questionnaire data. Most of conventional methods, however, aim at grasping overall trends in data, so they are not suitable for extracting and analyzing minority groups. EWOCS is one of the methods to extract minority groups, and isolated clique is the definition of groups similar to minority. However, EWOCS does not consider the dissimilarity to others and most of isolated cliques have a small number of data. In this paper, we define Local Minority Factor (LMF) as the criterion of minority to solve the above problems. LMF is based on Local Outlier Facor (LOF) which is one of the outlier detection methods and has two criterions as the isolation and the agglomeration. LMF becomes high when the data set is isolated but each data in the group is concentrated. We also propose an exploratory extraction method of minority groups by optimizing LMF. We apply the proposed method to an actual questionnaire data and compare with the conventional methods based on the criterion and the visualization. The results show that the proposed method can extract the respondents having stronger features of minority than the conventional methods and we show the examples of marketing strategy for each acquired minority group.
Focusing on word connectivity, we carried out an empirical analysis of vocabulary size changes with increasing sentences. While the regularity of a sublinear growth of vocabulary size with document size has been known, here we have found that the number of distinct connections between words and the number of types of connected wordswould grow in sublinear functions with document size.
In this paper, we formulate a simple recource programming problem for fuzzy random programming problems, in which equality constraints with fuzzy random variables areinvolved. In the proposed method, equality constraints with fuzzy random variables are defined through a possiblity measure. For a given permissible level of a possibility measure, the corresponding simple recource programming problem is formulated, which is called a fuzzy random simple recource programming problem. It is shown that a fuzzy random simple recource programming problem can be interpreted as a generalization of a simple recource programming problem.