Analysis of Best-Answer Estimation for a Q&A Site and its Application to Machine Learning

DOI IR Web Site Research Data 20 References

Bibliographic Information

Other Title
  • Ｑ＆Ａサイトにおけるベストアンサー推定の分析とその機械学習への応用
  • Q A サイト ニ オケル ベストアンサー スイテイ ノ ブンセキ ト ソノ キカイ ガクシュウ エ ノ オウヨウ

Search this article

Description

In this research, we investigated whether a computer could estimate the best answer on a Q&A site. First, a best answer estimation experiment was carried out with human assessors. The data of Yahoo! Chiebukuro was used for the experiment; 50 questions extracted at random from four categories, viz., "Consultation of love," "Personal computer,""General knowledge," and "Politics," were used. The accuracy rate(precision) of the estimation by two assessors was 50% and 52% (random estimation: 34%) for "Consultation of love," 62% and 58% (random estimation: 38%) for "Personal computer," 54% and 56% (random estimation: 37%) for "General knowledge," and 56% and 60% (random estimation:35.8%) for "Politics." Next, the experimental results were analyzed, and the machine learning system with "Detailed","Evidence", and "Polite" in the feature as a factor to choose the best answer was constructed. The precision of the machine learning system exceeded the assessors' results in the "Personal computer"(67%) category, and it fell below the assessors' results in the "Consultation of love"(41%) category. In the "General knowledge" and "Politics" categories, the precision of the machine learning system was almost equal to the assessors' results.

Journal

References(20)*help

See more

Keywords

Details 詳細情報について

Report a problem

Back to top