Kohei Otake

The COVID-19 epidemic has drastically changed our way of life. Especially in Japan, the lodging industry was hit hard by the trend toward self-restraint in travel. The situation is now returning to what it was before the epidemic, and businesses need to review their future facility operations. The purpose of this study is to typify and understand the characteristics of lodging facilities by focusing on their revenue management methods. Specifically, we use a questionnaire of employees involved in decision-making regarding the facility. First, we performed principal component analysis on 13 question items related to current profit management among all questionnaire items. We summarized the questionnaire items into 6 principal components by this analysis. For each principal component, we interpreted the first principal component as focusing on constancy, the second principal component as focusing on demand forecasting, the third principal component as focusing on room occupancy, the fourth principal component as focusing on customer demand, the fifth principal component as focusing on competitors, and the sixth principal component as focusing on company policy.Subsequently, we performed cluster analysis using the principal component scores obtained by principal component analysis. We calculated the average principal component score for each cluster, and named and discussed each cluster with reference to the calculated value.This study allowed us to develop a classification of facilities based on their revenue management methods.

In recent years, with the spread of COVID-19 infection, the consumer market in real places such as department stores and shopping centers was hit hard. The latest consumer trend surveys indicate that what consumers are looking for in physical shops after the coronavirus has been contained is 'confidence' and 'surprise' through the experience of touching actual products. From this, it can be inferred that the insight of consumers in actual shops lies in the process leading up to the purchase, i.e. the experience value, such as how they use the shop and search for products.In this study, we conducted an experiment on consumer behavior in a department store, and proposed marketing measures unique to a real shop in accordance with consumer behavior, using data on the flow line of shop movement and questionnaire data before and after the experiment. Specifically, a series of data on the consumer's flow line from entering to leaving the shop is collected using an eye-tracking device. From the collected traffic line data, we attempt to evaluate consumer behavior from the viewpoint of purchase purpose and loyalty by using Social Network Analysis(SNA) methods.

Since the early 2000s, the Internet has become increasingly popular for the development of information dissemination technology and as a platform for interaction. Therefore, the penetration rate of Social Networking Services (SNSs) is also increasing. Using the accounts created on SNSs, companies can disseminate information and communicate with users on SNSs for marketing purposes. Moreover, there are several influencer marketing activities that use influencers who are highly influential in their surroundings as marketing using SNSs. In this study, we aim to identify influencers on Twitter and consumer network structures for six cosmetic brands. Specifically, create a consumer network for each of the six cosmetic brands using follower data obtained from Twitter is created to identify the network structure. Furthermore, brand influencers were also identified. The consumer network of all six cosmetic brands was created to identify the influencers in the cosmetics industry. We compared the influencers of the brands with the influencers of the entire industry to examine any differences

The purpose of this study is to gain useful marketing insights by estimating the writer's emotions from the text and clarifying the relationship between the text's characteristics and food. First, each day’s emotional evaluation, taste evaluation, and the presence or absence of an event are clarified from the text, and each text is evaluated in three items. Next, the overall score is calculated from the average value of the three items, and the characteristics of sentences with high and low scores are clarified. The result of our analysis, the characteristics of the text reveal the relationship between emotions of the day and food choices.

In recent years, the number of beauty salons has been on the rise and competition is intensifying in Japan. It is necessary for stores to understand the characteristics of other stores in the area and the needs of their customers and to come up with strategies to secure customers to keep up with the competition. The purpose of this study is to understand the characteristics of stores using the review data. Specifically, we extract the characteristic words of each store and classify them according to their characteristics to understand the characteristics of beauty salons in the region from the review data using natural language processing. Furthermore, we use the extracted feature words to compare the feature of a target store with those of its competitors. Finally, we attempt to propose a marketing measure to the target store that considers the characteristics of local stores.

In recent years, the number of customers in physical stores has been declining because of the expansion of the EC market. Therefore, in physical stores, it is necessary to investigate effective product shelves and customers’ latent purchasing needs, which cannot be found only in purchase data to take advantage of the strengths of physical stores. The purpose of this study is to identify the golden zone which is attractive and easily gazed at by customers in an electronics retail store. In this study, we conducted an eye tracking observation experiment in an electronics retail store in Japan. From the experimental data, we aimed to obtain the subject’s movement lanes and viewpoint information. For the analysis, we used t-test to compare the differences in gazing time at the product shelves in different areas on the same floor and network analysis to visualize the purchasing behavior in a store. Based on the results of the network analysis, The area of interest (AOI) analysis was conducted on the product shelves with high degree centrality and betweenness centrality. The AOI analysis enables us to measure the number of gazes and gazing time of the area of interest by specifying the area of interest from the recorded data.

This study suggested the Deep Neural Network (DNN) model to estimate the ratio of reuse price to list price in the electrical industry, so that it can help to decide resale price. In reuse industry, the pricing is particularly important. First, we constructed one multiple regression model and two DNN models. The DNN models are divided according to the handling methods of categorical variables, One-hot vector and entity embedding. We introduced some explanatory variables in consideration the complicated consumer’s emotion for products price. Next, we utilized and evaluated these models. As the result, DNN with entity embedding is the best model based on correlation coefficient and RSME. However, we found that this model did not fit for the data of bulk buying. Lastly, we mentioned the problem of this model to improve accuracy.

Currently, competition for customers in the shoes industry is intensifying due to the entry from other industries such as the apparel industry. Therefore, we believe that it is an important issue for each brand to understand the values of its customers and to differentiate itself from other brands in order to build a strong relationship among them. In this study, we use a questionnaire data to understand the values of customers who own each shoes brand, and believe brands that have customers with similar values as brands that are likely to compete with each other. In addition, using a word-of-mouth review data of EC site, we identify differences in the directionality of each shoes brand based on the composition of word-of-mouth reviews for each brand. Based on these results and the demographic data of each brand owners obtained from the questionnaire data, we discuss the points that should be differentiated among some brands that are likely to compete with each other in terms of customer values.

In recent years, supermarkets and other retail stores have become increasingly competitive in their marketing as consumer behavior has diversified. In Japanese supermarkets, one of the major marketing strategies is event marketing. To measure differentiation among companies in the future, it is necessary to recommend products that are suitable for the needs of individual consumers. One of the methods used to understand consumer needs is Natural Language Processing using customers’ voices or review posts. By using Natural Language Processing, it is possible to estimate emotions that lead to consumers’ needs and decision-making. In this study, we focus on consumers’ emotions and aim to estimate complex emotions using diary data on purchasing behavior about dietary habits. Specifically, we created a new Japanese emotional expression dictionary by adding documents of the Japanese version of Wikipedia to the existing Japanese emotional expression dictionary. We then estimated the emotion of each text from the diary data using the dictionary. Furthermore, we conducted a questionnaire survey using subjects for the estimated emotions and evaluated the accuracy of the results.

Many people are using smartphone applications and websites with respect to cooking recipes in almost everyday life. By using cooking recipe sites, we easily search the cooking recipe for dishes that we want to eat and cook the dish based on the recipe. When we search cooking recipes, we refer to the two information that are pictures of cooked food and texts of cooking recipe. In this study, to investigate good cooking recipes, we built a multimodal coupled network to use these two kinds of information of cooking recipe sites. In this multimodal coupled network, as pre-learning, we learned images and texts separately. To learn texts, we first used word2vec to vectorize for each text, and then we use the deep averaging network to identify the objective function. To learn images, we performed transfer learning of images using VGG16 which is a famously trained model. We combined these two kinds of 300-dimensional feature vectors extracted from the final layer of the two models obtained by pre-learning, and then we learned a total of 600-dimensional feature vectors. As a result, the multimodal coupled network that combines image and text data has the best accuracy.

User Generated Contents (UGC) is one of representative tools to grow and make popular internet. In many UGCs, some influential users and groups exists, and various studies focused on these users. In recent years, “user-generated cooking recipe sites” have also been populated, and many consumers can post their own recipes. Many studies on recipe sites focus on evaluation of recipes, such as the number of evaluators and average scores. However, there are few studies focused on user relationships on cooking recipe site. In this study, we assume that cooking recipe sites have a user relationship structure similar to social media. Then, we aim to identify the relationship among users on cooking recipe site. Specifically, we use network analysis to visualize the relationship between recipe contributors and Reviewers on Japanese recipe sites. In addition, we used Modularity for community detection to identify similar groups and influential users.

In recent years, the popularity of e-commerce has witnessed a significant uptick. Physical apparel stores need to implement measures that focus on the behavioral experience of shopping at physical stores, a trait that e-commerce lacks. The purpose of this paper is to clarify the relationship between customer values and product search behavior and proposed product placement and customer service methods based on their values. We used questionnaire data on the values of customer purchasing to perform factor analysis and cluster analysis. Moreover, we extracted the product search behavior using eye-tracking gaze data from an apparel physical store. The results showed that product search behavior differed based on three types: trend cluster, self-esteem cluster, and conservative cluster. Finally, we proposed product placement in a store considering the features of these clusters.

In Japan, the number of beauty salons has been increased and reached over 250,000 at the end of 2018. On the other hand, there are some salon management problems such as price reduction and customer decrease. In such situations, it is an important not only to serve salon services, but to promote private brand (PB) items in addition to treatments. In this study, we analyzed purchase interval of PB items and predicted the purchase in a hair salon chain. First, we created explanatory variables using ID-POS data of the hair salon chain. Secondly, we selected explanatory variables using Cox proportional hazard model. Then, we performed Bayesian survival analysis to evaluate purchase interval considering customers’ heterogeneity. As a result, we could grasp appropriate timing when customers had highly purchase motivation and applied marketing measures to effective promotion.

Over the last few decades, consumers’ preferences and lifestyles have changed significantly. In such market, the number of products and services aims to mass marketing have become less effective. Therefore, personalization tailored to individual characteristics, certain segments, and one-to-one marketing focusing on individuals are becoming important. Therefore, customer relationship management, retaining existing customers and costs for acquiring new customers are important for both of academic and business field. However, for that purpose, it is essential to grasp customer behavior in more detail and use it for analysis. Therefore, in this study, we estimate the potential clusters of customers and customers’ purchase behavior. Concretely, we use pLSA and XGBoost which become popular machine learning methodologies. In this study, we set the number of store visits per month for objective variable and show a prediction model of it. Then we compare the model that incorporates the result of predicting the probability of the latent cluster as an explanatory variable with the model that incorporates a general explanatory variable.

pLSA is a useful method to know the characteristics of customer or item in marketing. In this study, we proposed a method to set the initial values more efficiently than the existing method for the problem that the final solution depends on the initial values set in the EM algorithm used by pLSA to estimate the solutions. We focused on the dimensional compression and clustering that are the characteristics of pLSA, and thought that the stability of the solution of pLSA would be improved by reflecting it in the initial values. Therefore, first, we performed correspondence analysis and k-means cluster analysis on the original data to express the features of dimensional compression and clustering. Next, we compared the performance of the pLSA results with the initial values of the proposed method and the initial values of the conventional method using random numbers. As a result, it was shown that the proposed method also converges to the same log-likelihood as the conventional method, and that the proposed method is superior in terms of convergence speed and stability.

Social media services, including social networking services (SNSs) and microblogging services, are gaining prominence. SNSs have a variety of information on products and services, such as product introductions, utilization methods, and reviews. It is important for companies to utilize SNSs to understand the various ways of engaging with them. Against this backdrop, numerous studies have focused on marketing activities (e.g., consumer behavior and sales promotion) using information on the internet from sources such as SNSs, blogs, and news sites. In particular, to understand the dissemination of information on the Internet, various researchers have undertaken studies pertaining to the diffusion phenomenon occurring in the real world. Here, topic diffusion is a phenomenon whereby a certain topic is shared with several other users. In this study, we aimed to evaluate the diffusion phenomenon on Twitter. In particular, we focused on the state of a targeted topic and analyzed the estimation of the topic using natural language processing (NLP) and time series analysis. First, we collected tweets containing four titles of animation broadcasts using hashtags. Approximately 250,000 tweets were posted on Twitter in a month. Second, we used NLP methods such as morphological analysis and N-gram analysis to characterize the contents of each title. Third, using the time series data for the tweets, we created a mixture model that replicated the diffusion phenomenon. We clustered the diffusion phenomenon using this model. Finally, we combined the features related to the content of the tweets and the results of the clustering of the diffusion phenomenon and evaluated them.