以興趣關聯圖為基礎之推薦系統研究

Abstract

本研究應用社會網路的概念於推薦系統，將興趣之間的共有關係轉成類似社會網路的「興趣關聯圖」，然後以興趣關聯圖上不同興趣之間的連結強度作為以興趣推薦興趣的依據。本研究的主要目的為確認此種以興趣關聯圖為基礎的推薦系統的可行性，同時比較此方法在即時性與動態系統上相對於一般推薦系統的優勢。 當某人同時喜愛兩個興趣，則二興趣間即產生關聯；透過多組興趣與多人的搭配重複此關聯動作，由此過程建立興趣關聯圖。利用興趣間的關聯強度給予使用者推薦之後，一方面計算擊中興趣個數占推薦興趣個數的比例，此值表示關聯圖預測興趣的精確程度；二方面則計算擊中興趣個數占使用者實際選擇興趣個數的比例，此值可反應使用者回應推薦系統的程度、即回應率；經衡量精確程度與回應率，本研究以精確程度與回應率來判斷系統之可行性。 本研究認為以興趣關聯圖為基礎的推薦系統其即時性可透過動態系統演算法達成，經此過程而解決一般推薦系統需定時更新推薦內容之疑慮，隨時更新系統推薦伴隨的優點則是使用者因此而得到最新推薦，或可提升推薦系統品質與使用者之信任。其次影響的是改善推薦系統效率，動態系統允許時時刻刻檢視系統內部興趣關聯圖的現況與更新，使得推薦系統既可時時處於運算狀態、同時也可視為時時處於備妥狀態，只等使用者輸入嗜好即可判別所應輸出之結果。 根據以上理由，本研究提出的以興趣關聯圖為基礎的推薦系統，的確可行且較傳統推薦系統有其局部優勢。

By applying the concept of social network into recommendation system, we convert the relationships between interests into ‘Interest Map’, just as the social network looks like. According to the association strength, the system could recommend users interests from interests. The goal of study is to verify if the recommendation system based on Interest Map is feasible, and to compare the relative advantages of immediate computation, and dynamic system over the general recommendation systems. The relationship between two interests, here we call it association, is built when someone likes these two interests at the same time. Repeating the process of association-building, we make Interest Map. After recommendation, which is selected from the strongest strength of associations, we compute the precision rate and recall rate to verify if the recommendation system based on Interest Map is feadible. Our study suggests that the feature of immediate computation is achieved by the dynamic algorithm, and meets the need of routine update of the general recommendation systems. By this process, users could get the newest recommendation at any time, and may enhance the recommendation and user trust. Besides, dynamic system improves the efficiency of recommendation system. The feature of dynamic system allows the recommendation system to check the Interest Map inside and update in time, and makes the recommendation system at a prepared condition to response users’ request. Owing to the reasons above, the recommendation system based on Interest Map is feasible and has some relative advantages over the general recommendation systems.