トピックモデルを用いた橋梁点検結果所見記述の維持管理への活用に関する基礎的分析

抄録

Utilization of collected and accumulated data of the bridge inspection results for maintenance is required. Although the qualitative evaluation items of the bridge have previously been often used, in this research the opinion of bridge inspection by technical expert is focused. Topic model is probabilistic modeling of term frequency occurrences in documents, which can be used to classify documents of the opinion of bridge inspection. And apply a topic model for the text data to detect the characteristics of bridge damage not recorded as database evaluation items. The topics corresponding to the characteristic damage condition of the bridge are organized. A topic has been organized by high probability words about deterioration due to ASR (Alkali-Silica Reaction), which is a characteristic deterioration in the Hokuriku region. Focusing on the opinion of bridge associated with high probability topic, it is possible to do early detection and preventive maintenance. While these effective results were obtained, some topics were organized by characteristic writing by the inspection company. This means that in order to accumulate the inspection data while maintaining quality, it is necessary to decide the criteria of opinion method of listing and contents.