With progress of information technology, vocal sounds have been able to edit by software. However, it is still difficult for beginner users to control vocal quality with reflecting users' intention. This study proposed a method that optimizes vocal sound suited to each user's intention by using Interactive Genetic Algorithm (IGA). IGA is an interactive type of Genetic Algorithm (GA), which is one of evolutionary algorithms searching optimal solutions. IGA employs user's subjective evaluation as fitness function in GA. Target of the proposed IGA was to find better and best parameters of vocal quality. Listening experiments were conducted to investigate the fundamental efficacy of the proposed IGA. Eleven subjects evaluated vocal sounds presented from a system based on the proposed IGA. To create vocal sounds, VOCALOID, software synthesizes vocal sounds was used. Results of the experiments showed that the proposed IGA successfully searched better parameters of vocal quality.
Increased demand for web advertising has resulted in a corresponding increase in the need to develop online personalized advertisement. This paper proposes an advertising slogan generation system reflecting preferences of users on the web. By using a social networking service (SNS) site as knowledge base of word preferences, and by employing an advertising slogan corpus, the proposed system aims to generate slogans that reflect advertising posts on SNS. Using model slogans selected from the corpus containing 24,472 slogans, the proposed system generates slogan candidates using the knowledge obtained from a post on SNS. These slogan candidates are selected by the following three indexes; the natural level given by a large scale balanced corpus, the semantic relations score using the advertising slogans, and the preference level obtained from SNS sites. Especially, the proposed system extracts preference data from these SNS fan pages and estimates preference level on each word based on bag-of-words model. This enables the proposed system to select slogans in fashion. The authors conducted an objective experiment to examine the quality of generated slogans. The result shows that (1) the natural level and semantic relation level are effective to select slogans that reflect a post (2) the preference level index contributes to select preferred slogans that interest people.
In this paper, we propose an automatic generation system of the comic from a newspaper article. In the proposed system, two characters appear in the generated comic: one has a role of questioner and the other is elucidator. These characters are created using an automatic comics generation tool named “Komipo!”. First, the proposed system analyzes the text of a newspaper article and every sentence is presumed the importance and importance analysis and emotion analysis using language resources are carried out for each sentence. From the result of importance and emotion analysis, the system chooses the characters having suitable expression and a pose to the sentence of an article. The system converts the text of an article into spoken language from the literary language. If there is a difficult word in a text, the proposed system generates additional frame to explain the meaning of the difficult word. The background image for every frame is chosen from Flickr. In the image selection process, not only language processing, but also the result of image analyses is used. Evaluation experiments have been carried out and we confirmed that the system is able to choose the image suitable for the text of an article by performing image processing. In addition, it has shown that the expression by a comic style contributes to readability and easiness of understanding for newspaper articles.
A method for modeling a tacit knowledge has been developed. Tacit knowledge is a type of knowledge that is difficult to describe with words or symbols such as sports techniques, design skills, etc. It is difficult for us to obtain and share this type of knowledge. In this paper, we discuss a new method to create a model of tacit knowledge by showing an example of web site designing. Usually, capable designers can express their intensions precisely through their works. This is to say, appearances of their works reflect their intention. Thus, it is expected that capable designers' tacit knowledge can be extracted from their works. Knowledge of the design was extracted from actual web pages on the Internet. Then, knowledge models were created by utilizing a Bayesian network. The created models were verified by LOOCV technique. The result shows that one of the created models illustrates part of a designers' knowledge accurately.
In this paper, we propose a dialogue system with knowledge acquisition ability from the user's utterances. The following two points are considered in this system. First, the user's preference is estimated from user's utterances. In the proposed system, each user's information such as the utterances, the topic words and the deep cases are memorized. The proposed system estimates user's preference using the emotion estimation and the topic words focusing on verbs. Also, these links can be learned based on the co-occurrence frequencies of emotion estimation. Second, the user's preference is utilized the dialogue system. The proposed system can select a more suitable topic using a user's preference. In addition, the proposed system can make use of the other person's utterances. Evaluation experiments have been carried out and we confirmed that the proposed system could improve the satisfaction level by usage of the stored information by the user. In addition, we evaluated the proposed system by elderly peoples. We confirmed that the proposed system is effective also as an attentive hearing system for elderly.
Many findings of behavioral decision research suggested that attention had a crucial role in decision-making. We proposed a method to induce attention toward particular attribute such as money using control by coefficient of variation in multi-attribute decision task. The experiment consisted of two steps: 1) arousing attention using the proposed method and 2) testing the effect of the method. We hypothesized that the participants would pay more attention to high, relative to low, CV attributes. The results for the first step of the experiment showed that the participants paid more attention to high, relative to low CV probability and low, relative to high, CV outcome. Results for the second step of the experiment were similar to those observed in the first step. This finding indicates that the proposed method was effective in arousing attention during the decision-making process.
We examined the effect of discount rate presentation (e.g. 40% off) on the consumers' decision making process using eye movement equipment. Ten female participants (average age: 48.4 years old) were asked to choose between two options. The discount rate of option one was higher than that of option two, although total price of option one was higher than that of option two. All eye movements were recorded during experiment in order to examine participants' decision making process. To recheck this finding, we conducted a control experiment without discount rates. The result of the control experiment suggested that the presentation of discount rate influenced the participants' decision strategy.