臺大管理論叢 NTU Management Review VOL.28 NO.3

93 NTU Management Review Vol. 28 No. 3 Dec. 2018 The expansion of concept “1.1.2 Design Styles” can be used as an example. Assume that the initial activation values (i.e., the interest level of the concept as rated by the focal user) of concepts “1.1.2 Design Styles” and “1.2.1 Design Styles” are 4 and 6, respectively. The concepts expanded at different spreading distances are shown in Table 2, along with their respective activation values. Concepts whose activation values are higher than the spreading threshold are used for assessing documents that are to be recommended at the document recommendation phase. Table 2 Example of Expansion of Concept “1.1.2 Design Styles” Spreading Distance Expanded Concept Activation Value 1 1.1 Memory Structures (0) 0+4×0.7 = 2.8 1.1.2.1 Primary memory (0) 0+4×0.4 = 1.6 1.2.1 Design Styles (6)* 6+4×0.5 = 8 2 1. Hardware (0) 0+2.8×0.7 = 1.96 1.1.1 Semiconductor Memories (0) 0+2.8×0.4 = 1.12 1.2 Logic Design (0) 0+8×0.7 = 5.6 Note: *The number in parentheses represents the initial activation value of the concept. 3.3 Document Recommendation On the basis of the original concepts and the array of expanded concepts, the ICE technique evaluates all the documents and recommends those that may appeal to the focal user. Specifically, the ICE technique calculates the “interest score” for each document by summing up the activation values of the concepts existing within it. For example, assume that a document contains the “1.2.1 Design Styles,” “1.1.2 Design Aids,” and “1.1.2.1 Primary memory” concepts. As shown in Table 2, the activation values of these concepts are 8, 0, and 1.6, respectively. The document, as a result, obtains an interest score of 8+0+1.6 = 9.6. Finally, the top- n documents that achieve the highest interest scores are recommended to the focal user. 4. Experimental Evaluation Results A laboratory experiment was conducted to compare the system performance of ICE with three benchmark techniques: Explicit-feedback-based Concept-Expansion (ECE), keyword-based, and random-based document recommendation. An experimental online periodical database was designed to provide four kinds of recommendation services. The ACM Computing Classification System (CCS) was employed as a domain-specific