Michiharu Yamashita

Michiharu Yamashita

PhD Candidate at Penn State

Portfolio

BigGorilla

BigGorilla is open-source components for data integration and preparation, which began in 2016 jointly by Recruit and University of Wisconsin at Madison. It documents existing technologies and our original technologies to solve the problem. I created a couple of components of BigGorilla, and evangelized them. I also applied these technologies into 8 companies within Recruit, and showed that BigGorilla is effective across the company’s diverse range of businesses: the extraction of store names (or person names and location information) from unstructured data, merging of lists from multiple data sources, etc. For example, with BigGorilla, we obtained 98.9% accuracy on the task of de-duplicating approximately 10,000 store names (Here is the press release at that time).

Spysee2

Spysee2 is one of the biggest Japanese search engine of people and their relationship on the web. I extracted a person’s social network from the web using heuristic techniques and developed contemporary machine learning and NLP based algorithms for the extraction of various types of information about people (bio-data, profile picture, social network, related web pages etc.) from unstructured and noisy web data. I also developed a solution for name disambiguation problems in short texts using a probabilistic classifier. (Here is the press release, the article, and the blog at that time).

Pedestrian Congestion Visualization

Developed a pedestrian congestion visualization algorithm using GPS based users’ log and OpenStreetMap’s road information, which enables companies to analyze the time and day when area that they want to know will be crowded and to check how the road is congested on the map.

Smartphone Application’s Competitiveness Visualization

Developed a smartphone apps competitiveness visualization algorithm and approach, which visualize the competitivenes of apps and how the app will survive on smartphones based on how long users possess their applications. (This work was collaborated with Fuller, which have the biggest service of mobile app market analytics in Japan, App Ape. Recently, this technology was released as the real application and here is the press release.).