企業知識文件推薦系統建置之研究

Abstract

現今企業環境中，透過各種Content Management System (CMS)，已能達到企業內各種知識的收集、整理、保存等功能。但在知識的分享上，仍是要透過繁雜的關鍵字搜尋(被動式的知識分享方式)；而成熟的CMS，又包含了過多的資訊，往往造成知識分享的效果不好。故希望建置一個知識推薦系統，透過主動式的分享，達成企業內知識的傳承。

本論文研究現有的推薦方法、主題模型技術，找出適用於企業知識推薦系統的方法，建置適用於企業內知識分享的推薦系統，以促進企業知識分享。本研究建置三種推薦模型，並透過實驗分析隱含主題LDA方法之適合主題數目，以及考慮文件查閱數來優化推薦方法，實驗結果顯示矩陣分解法(MF)結合LDA的推薦方法優於其它方法。

In today's business environment, enterprises have been able to collect, organize and preserve all kinds of knowledge through a variety of Content Management System (CMS). However knowledge sharing is still to be complicated by keyword search (passive knowledge-sharing approach); and mature CMS often contains information overload, which results in ineffective knowledge sharing. Thus, it is important to build a knowledge recommendation system through active sharing within enterprises. This study investigates the existing recommendations, topic model techniques, to find out suitable recommendation methods for enterprise knowledge management systems. An enterprise knowledge recommender system is developed in order to achieve the promotion of knowledge sharing within the enterprises. The study designs three recommendation models and conducts experiments to discover suitable number of topics for LDA approach. Experimental result shows that Matrix Factorization (MF) combined with LDA approach performs better than other methods.