Tang Q Zeng/The journal of Systems and Software 85 (2012)87-101 COMPUTER SICENCE COMPUTER SICENCE Algorithms and data Algorithms and data tructures neural network Artificial intelligence Fig. 6. The result of the ontology extension. refers to the users interest in a certain topic about which no papers both keyword and keyword. where paper E UPS(user. have yet been directly accessed by the user. Through the ontolog UPS(user) represents the set of papers recorde ical relationships between this topic and the other topics in which behavior history data; keyword, keyword E UTG(topic usern s he user has explicit interests, the users implicit interest about the weights of the edges are defined as the number of related nis"uninteresting"topic can be predicted Implicit interest pro- files express users' potential interests in topics that are obscurely (4) The weight of node keyword in UtG which is referred as (keyword) measuring the user's interest in the content Definition 3(User interest profile). A user interest profile consists related to keyword; is calculated by of two parts: an explicit interest profile part and an implicit interest profile part, denoted by UEIPUseru)and UlIPUseTu), respectively Each of the two parts is a set of 2-tuples: UEIP(Useru)-( topic. EI( IK(keyword, )=> (WKP( eyword, paper).UIP(paper, user). topic, user ))ltopicr is a topic, El( topic, user)measures User: explicit interest in topic): and UIIPUseru=[( topick, ll( topick, where keyword; E KNS: PS(keyword. seru))ltopick is a topic, Il( topick, user)measures Useru's implicit set of papers con- interest in topick) are ups use s u): UPS(user )is the 5. 1. The profiling method for users' explicit interests WKP(keyword, paper)is the weight of keyword in the vector space model of paper And We now introduce the construction of the explicit interest pre file. The generation of explicit interest profiles is achieved through UIP(paper, user)=-BF(paper the user profiling module in Fig. 1. Fig. 7 shows a detailed process DP( paper,lse%,Φe(0,+∞) i> The key task of creating the explicit interest profiles of users lies number of times user browses paper, or the number of timese he chart of this module where BF(paper user)denotes behavior factor which equal nave explicit interest. Considering a paper's possible relevance to depending on the type of behavior based on whhs divided by 5 i computing the interest values of the topics in which the users downloads paper, or the score user rates paper different topics, the possible relevance of a keyword to different culated: DP(paperp, user )indicates the number of days that have topics is taken into account. In order to differentiate between the passed since the behavior occurred; p is a parameter for adjust pics for which a user has the same access behaviors, we assign a ment neasurement to each topic called the relevance factor. In the fol- d should be set as a value which optimizes the recommenda lowing, we first group of algorithms and definitions, and then tion performance. In the conventional ontology-based user interest ve present the definition and explanation of the relevance factor profiling algorithm, is set as 1(Middleton et al, 2004) and the explicit interest profiler. Definition 4(Inneredge ). An edge is termed an inner edge if it links Algorithm 1( Generating user-featured topic graph). Based on the two keyword nodes in the same weighted keyword definition of weighted keyword graph WKG, a user-featured topi graph, which is referred as UIG, can be created in the following edge strength. In this paper, we use IES(keyword, de iscalled to the inner edge strength of keyword, (1)Create the weighted keyword graph WKG(topic)for topic Definition 5 (Cross edge). An edge is termed as a cross edge if it (2) Eliminate all edges and reset weights of les to o to gen- links two nodes from different weighted keyword graphs. The sum erate a new graph UTG( topic). of the weights of the cross edges of a node is called cross edge (3)Based on the user's behavior history data, assign an edge strength. In is paper, we use CES(keyword) to refer to the cross tween keyword and keyword only if paper contains edge strength of keyword,X. Tang, Q. Zeng / The Journal of Systems and Software 85 (2012) 87–101 93 Fig. 6. The result of the ontology extension. refers to the user’s interest in a certain topic about which no papers have yet been directly accessed by the user. Through the ontolog￾ical relationships between this topic and the other topics in which the user has explicit interests, the user’s implicit interest about this “uninteresting” topic can be predicted. Implicit interest pro- files express users’ potential interests in topics that are obscurely discovered. Definition 3 (User interest profile). A user interest profile consists of two parts: an explicit interest profile part and an implicit interest profile part, denoted by UEIP(Useru) and UIIP(Useru), respectively. Each of the two parts is a set of 2-tuples: UEIP(Useru) = {(topict, EI( topict, useru))|topict is a topic, EI( topict, useru) measures Useru ’ s explicit interest in topict}; and UIIP(Useru) = {(topic, II( topic, useru))|topic is a topic, II( topic, useru) measures Useru ’ s implicit interest in topic}. 5.1. The profiling method for users’ explicit interests We now introduce the construction of the explicit interest pro- file. The generation of explicit interest profiles is achieved through the user profiling module in Fig. 1. Fig. 7 shows a detailed process chart of this module. The key task of creating the explicit interest profiles of users lies in computing the interest values of the topics in which the users have explicit interest. Considering a paper’s possible relevance to different topics, the possible relevance of a keyword to different topics is taken into account. In order to differentiate between the topics for which a user has the same access behaviors, we assign a measurement to each topic called the relevance factor. In the fol￾lowing, we first give a group of algorithms and definitions, and then we present the definition and explanation of the relevance factor and the explicit interest profiler. Algorithm 1 (Generating user-featured topic graph). Based on the definition of weighted keyword graph WKG, a user-featured topic graph, which is referred as UTG, can be created in the following steps: (1) Create the weighted keyword graph WKG(topict) for topict. (2) Eliminate all edges and reset weights of all nodes to 0 to gen￾erate a new graph UTG(topict). (3) Based on the user’s behavior history data, assign an edge between keywordi and keywordj only if paperp contains both keywordi and keywordj, where paperp ∈ UPS(useru) and UPS(useru) represents the set of papers recorded in useru’s behavior history data; keywordi, keywordj ∈ UTG(topict). And the weights of the edges are defined as the number of related papers. (4) The weight of node keywordi in UTG which is referred as IK(keywordi) measuring the user’s interest in the content related to keywordi is calculated by IK(keywordi) =  paperp ∈ PS(keywordi) (WKP(keywordi, paperp) · UIP(paperp, useru)), where keywordi ∈ KNS; PS(keywordi) is the set of papers con￾taining keywordi and PS(keywordi) ∈ UPS(useru); UPS(useru) is the set of papers recorded in useru’s behavior history database; WKP(keywordi, paperp) is the weight of keywordi in the vector space model of paperp. And UIP(paperp, useru) = BF(paperp, useru) (DP(paperp, useru))1/˚ , ˚ ∈ (0, +∞), where BF(paperp, useru) denotes behavior factor which equals the number of times useru browses paperp, or the number of times useru downloads paperp, or the score useru rates paperp divided by 5, depending on the type of behavior based on which TG(topict) is cal￾culated; DP(paperp, useru) indicates the number of days that have passed since the behavior occurred; ˚ is a parameter for adjust￾ment. ˚ should be set as a value which optimizes the recommenda￾tion performance. In the conventional ontology-based user interest profiling algorithm, ˚ is set as 1 (Middleton et al., 2004). Definition 4 (Inner edge). An edge is termed an inner edge if it links two keyword nodes in the same weighted keyword graph. The sum of the weights of the inner edges of a keyword node is called inner edge strength. In this paper, we use IES(keywordi) to refer to the inner edge strength of keywordi. Definition 5 (Cross edge). An edge is termed as a cross edge if it links two nodes from different weighted keyword graphs. The sum of the weights of the cross edges of a node is called cross edge strength. In this paper, we use CES(keywordi) to refer to the cross edge strength of keywordi.