近年來，越來越頻繁的相關新聞報導機器人或是電腦等即將出現在我們生活周遭。但是要讓電腦像人類一般靠著生活周遭的事物去學習到人類的基本常識，仍然有其困難度。除此之外，如何用規則定義這些常識也是另一件不容易的事情。常識推理的模式有很多種，而聯想推理是常識推理學習的基本能力之一。本論文以常識聯想的推理與學習為主要研究標的。
此篇論文中納入了基於基本常識建立的ConceptNet知識庫作為學習常識的特徵， 並且利用非負矩陣分解(Non-negative Matrix Factorization，NMF)方法進行訓練聯想學習。論文中的實驗採用兩組資料，667組地點-活動和625組目標-行動聯想的資料當作訓練以及測試的資料集，並且使用交叉驗證（cross-validation），作為訓練機器學習的學習資料以及結果的評估。在地點-活動的資料，我們實驗結果得到準確率(Precision)約達77.2%，召回率（recall）約達31.2%，Ｆ分數達到42.3%，另一組資料則得到準確率(Precision)約達93.5%，召回率（recall）約達25.3%，Ｆ分數達到39.5%。
為了瞭解觀念聯想學習中特徵與特徵的聯結關係對於學習效益的影響，我們另外提出六種特徵選取策略以了解其對聯想學習的影響，我們使用的方法包含random、most-link (rank)、entropy、Singular Value Decomposition(SVD)，以及H-link再分為H-frequency和H-weight。
在實驗數據中發現，經過挑選後的特徵所得到的F分數和召回率(recall)都會優於最基本的隨機取樣（random）方法，也表示透過特徵選取可以得到較有意義的特徵去學習，並且提高Ｆ分數和召回率(recall)。
另外，兩組資料本身的性質也會產生不一樣的結果，我們在第三章討論。

Nowadays, more and more news and reports about a robot or a computer will appear in our life. It seems that the information technology is really close to us. However, even now we have the good development of information technology, It is still difficult for computers to learn the common sense possessed by mainkind via things encountered in the daily life as a human. Besides, it is non-trivial to formulate all common sense in terms of definite rules. There are many models of common sense reasoning, and the association reasoning is a fundamental ability of common sense. We focus on the learning on the common sense association reasoning.

We develop a computational model to learn the common sense association between a pair of concept classes based on a bipartite network and matrix factorization methods. In this model, we view the concept-pair association as a bipartite network so that the auto-association mappings can become similarity constraints. We impose the additional similarity and regularity constraints on the optimization objectives so that a mapping matrix can be found in the matrix factorization to best fit the observation data. We extract 667 location-activity pairs and 625 goal-action pairs from ConceptNet [12] as our training data and test data. We evaluate the performance in terms of F-factor, precision and recall using a common sense association problem between locations and activities against six feature selection strategies in the matrix factorization optimization. We reach performance of precision up to 77.2%, recall up to 31.2% and F-score up to 42.3% in the location-activity association domain. In the goal-activity association domain, we reach performance of precision up to 93.5%, recall up to 25.3%, and F-score up to 39.5%. For understanding the feature selection effects on the association learning, we assumed six feature selection methods including random, most-link (rank), entropy, singular value decomposition (SVD), H-frequency, and H-weight. We found that all performances of recall and F-score in the five feature selection methods are better than the performance of recall and F-score in the random method.

摘要 I
Abstract III
List Of Tables VI
List of Figures VII
1 Introduction and Related work 1
2 Methodology 4
2.1 Data Set Collection 4
2.1.1 Common Sense Pair Collection 5
2.1.2 Feature Collection 6
2.2 Methods 7
2.2.1 Objective 7
2.2.2 Material 8
1. Ground Truth Data 8
2. Feature Set 9
3. Similarity Measure 10
4. Bipartite Projection 11
2.2.3 Optimization Of Matrix H 12
2.2.4 Feature Selection Methods 14
3 Experiments and Discussion 17
3.1 Experiment 1 17
3.2 Experiment 2 18
3.3 Experiment 3 20
3.3.1 Location-Activity Association Domain 20
3.3.2 Goal-Action Association Domain 23
3.4 Experiment 4 25
3.4.1 Location-Activity Domain 25
3.4.2 Goal-Action Domain 28
3.5 Discussion 31
4 Conclusion and Future work 36
5 Reference 38

1. G. Brewka, Nonmonotonic Reasoning: Logical Foundations of Commonsense, Cambridge University Press, (1991).
2. Yumi Iwasaki, Real-world applications of qualitative reasoning, IEEE Expert: Intelligent Systems. Knowledge Systems Laboratory, Department of Computer Science: Stanford University, (1997).
3. Johan De Kleer, John Seely Brown, A qualitative physics based on confluences, Artificial Intelligence,Vol. 24, Issues 1–3, Pages 7-83, (1984).
4. Steven A. Sloman, The empirical case for two systems of reasoning, Psychological Bulletin, Vol. 119, No. 1, 3-22, (1996).
5. Jehoshua Bruck, On the convergence properties of the hopfield model. Proceedings of the IEEE, 78(10), October 1990.
6. Bart Kosko, Bidirectional associative memory, IEEE trans. On Systems, Man and Cybernetics, Vol. 18, Nov. 1, 1988.
7. C.-Y. Liou and S.-K.Yuan, Error tolerant associative memory, Biological Cybernetics, 81: 331–342, (1999).
8. Yehuda Koren, Robert Bell and Chris Volinsky, Matrix factorization techniques for recommender systems, IEEE Computer, Volume 42 Issue 8, 30-37, (2009).
9. Daniel D. Lee and Sebastian Seung, Learning the parts of objects by non-negative matrix factorization, Nature, Vol. 401, 788-791, (1999).
10. Tzu-Chun Chen and Von-Wun Soo, Learning common sense associations based on a bipartite network using matrix factorization, In proc. of ISIS 2015, Korea.
11. Feature selection, https://en.wikipedia.org/wiki/Feature_selection.
12. ConceptNet, http://conceptnet5.media.mit.edu/
13. Entropy(Information Theory), https://en.wikipedia.org/wiki/Entropy_(information_theory)
14. Singular Value Decomposition, https://en.wikipedia.org/wiki/Singular_value_decomposition
15. Kuang-Chin Hsieh and Von-Wun Soo, Drug target prediction based on similarity in chemical and genomic bipartite networks, Department of Computer Science, National Tsing Hua University, 2014.
16. Cosine similarity, https://en.wikipedia.org/wiki/Cosine_similarity
17. Bipartite graph, https://en.wikipedia.org/wiki/Bipartite_graph
18. Non-negative matrix factorization, https://en.wikipedia.org/wiki/Non-negative_matrix_factorization
19. Open mind common sense, https://en.wikipedia.org/wiki/Open_Mind_Common_Sense