Research and Development in AI-Driven Data Analysis and Business Process Automation

「This fiscal year, based on the issues identified with the conventional method, we focused on researching and developing a new inference model using generative AI. Previously, we vectorized product names and descriptions and estimated the optimal category based on semantic similarity, but identified the issue that accuracy did not reach a practical level when hierarchical structures differed among multiple marketplaces. In the consolidated fiscal year, we therefore developed a new script that inputs product names and the category hierarchy information for each marketplace into generative AI and has it infer the optimal category. As a result, accuracy improved substantially compared with the conventional method, and we established an operating method that controls the inference cost of generative AI. Following trial operation by the internal support team, the script has now moved to the stage where the development team is considering incorporating it into the product.」

「In both businesses, we view advances in AI technology as an opportunity to accelerate business growth.」

「This technology has already been developed into a tool and is now in practical operation in the company's internal support functions. In practical operation, it has achieved a high accuracy rate in the high 80% range for automatically linking attribute data, contributing to a substantial improvement in work efficiency.」

「In verifying inquiry support using the AI language model "GPT-4o mini," we generated responses based on the CROSSMALL support site and other related documents and evaluated them against actual inquiries. For product-name matching, we used BERTScore and STS (Semantic Textual Similarity) and conducted verification while working to improve accuracy and processing speed by fine-tuning the trained model and adjusting data weighting.」

「There were no material changes in the status of the Group's research and development activities during the cumulative first quarter.」

「Verification of the large language model (LLM) is nearly complete, and we are moving to verification of its implementation in existing services using a library that supports the development of applications integrated with LLMs. Specifically, while seeking to improve UI/UX through uses such as generating responses to inquiries, extracting similar product names and attributes from an entered product name, and checking the consistency of address data when placing orders, we plan to proceed with evaluation and verification focused primarily on accuracy and response time.」

「There were no material changes in the status of the Group's research and development activities during the cumulative second quarter.」

「There were no material changes in the status of the Group's research and development activities during the cumulative first quarter.」

「We had been working on use-case analysis for building interactive applications with the aim of partially automating inquiry-response operations. Based on the results of that analysis, we deployed conversational data in an external fully managed artificial intelligence service environment and proceeded with verification to optimize the Q&A. We plan to continue improving accuracy by using related functions as well.」

「There were no material changes in the status of the Group's research and development activities during the cumulative third quarter.」

「During the consolidated fiscal year, we worked on analyzing use cases for building interactive applications using voice and text, with the aim of partially automating inquiry-response operations. We plan to continue verification toward practical implementation based on our accumulated expertise in natural language processing, including the use of external cloud services. As a result of these initiatives, the group's total research and development expenditure for the consolidated fiscal year was ¥37.048 million.」

「The group's total research and development expenditure for the first nine months of the third quarter of the consolidated fiscal year was ¥28.633 million. There were no material changes in the status of the group's research and development activities during that period.」

「There were no material changes in the status of the Group's research and development activities during the cumulative first quarter.」

「During the consolidated fiscal year, with the aim of putting the "text recognition function" into practical operation, we enhanced a method for efficiently extracting strings obtained through the text recognition function from among groups of strings of correct labels (words). By combining multiple algorithms for calculating the similarity between strings in natural language processing, we developed an algorithm for evaluating similarity between strings in a way that more closely reflects human perception. We plan to continue speeding up processing while pursuing overall optimization. As a result of these initiatives, the group's total research and development expenditure for the consolidated fiscal year was ¥40.804 million.」

「There were no significant changes in the status of research and development activities of our group during the cumulative third quarter consolidated accounting period.」

「The group's total research and development expenditure for the first six months of the second quarter of the consolidated fiscal year was ¥20.899 million. There were no material changes in the status of the group's research and development activities during that period.」

「During the consolidated fiscal year, we worked on developing technology to interpret features in images that are close to the information obtained through human visual inspection by incorporating machine learning and deep learning alongside conventional image processing to identify latent patterns in images. Based on these results, we are proceeding with application to cloud services as a core function of a text recognition engine.」

「During the consolidated fiscal year, we focused primarily on feature engineering and worked on developing a general-purpose feature extraction tool. We intend to use this tool to analyze accumulated big data efficiently and develop technology that enables business automation. In October 2017, we also opened the Isle Matsue Lab as a research and development center. As a result of these initiatives, the group's total research and development expenditure for the consolidated fiscal year was ¥63.952 million.」

「The group's total research and development expenditure for the consolidated fiscal year was ¥42.304 million. From the consolidated fiscal year, we also began research and development activities in the business systems field involving AI-based data analysis and business automation.」

「From the first quarter of the consolidated fiscal year, we began research and development activities in the business systems field involving AI-based data analysis and business automation.」

「From the first quarter of the consolidated fiscal year, we began research and development activities in the business systems field involving AI-based data analysis and business automation.」

「From the first quarter of the current consolidated fiscal year, we began research and development activities in the business systems field involving AI-based data analysis and business automation.」