a comprehensive computation, uncontrolled addition of all the related terms leads to the dilution of the profiles with noise or unrelated terms. This dilution may have negative implications on the computation process where the similarity in the noise may contribute to the similarity values between entities. It is therefore desirable to have control over the types of relationships to be considered during this spreading process A SET SPREADER (a)Set Spreading (b) Graph Spreading Fig 1: Two Schemes for Profile Spreading he weights of the new related terms are proportional to the weights of the original term as the weight wi of a term ti indicates the importance of the term within a user profile. However, during spreading the weights of the related terms should differ accord- ing to the semantics of the relationships on the edge. For example, spreading based on wikipedia may be limited to only spreading along the parent categories. We therefore use a set of linear influence functions, one per relationship-type(role/property of an ontology), to control the spreading process. For example, a spreading process based on Wordnet limited to types synonym and antonym can have functions tij=wi X 0.9 and ti;=Wix-09 respectively. We propose two schemes for representing the related terms post-spreading: extended set and semantic network. 4.1 Set Spreading Depicted in Figure la, set spreading is a process of extending an user profile such that the related terms, which are determined with respect to an ontology, are just appended to the original set of terms. Set spreading is an iterative process. After each iteration, the related terms from the previous iterations are appended to the profile. The spreading process is terminated if there are no related terms to spread the profile with or after a fixed number of iterations 4.2 Graph spreading Shown in Figure 1b, graph spreading is the process where terms from two profiles and the related terms are build into a semantic network(SAN). Unlike set spreading, graph spreading preserves the relationship between a term in a profile and its related term in the form of a graph edge. This allows consideration of relationships based on their semantics on the same network. Graph spreading terminates like set spreading, or if there exists a path between every pair of the term nodes from the two profiles. This condition best uits the ontologies that have a top root element which subsumes the rest of the elementsa comprehensive computation, uncontrolled addition of all the related terms leads to the dilution of the profiles with noise or unrelated terms. This dilution may have negative implications on the computation process where the similarity in the noise may contribute to the similarity values between entities. It is therefore desirable to have control over the types of relationships to be considered during this spreading process. t1 w1 t2 w2 t3 w3 Ontology SET SPREADER # Types of relationships to spread upon (such as parentOf) # Weight functions for the spreaded terms # Termination condition (such as no. of iterations) Parameters t1r1 fr(w1) t2r1 fr(w2) t2s2 fs(w2) t1 w1 t2 w2 t3 w3 t3r1 fr(w3) t3r2 fr(w3) t3s1 fs(w3) # Iteration = 1 # Relationships = {r,s} # Weight functions = {fr,fs} (a) Set Spreading t11 w11 t12 w12 Ontology GRAPH SPREADER # Types of relationships to spread upon (such as parentOf) # Weight functions for the spreaded terms # Termination condition (such as no. of iterations) Parameters # Relationships = {r,s} # Weight functions = {fr,fs} t21 w21 t22 w22 Profile 1 Profile 2 t11 t12 t21 t22 t11r1 t12s1 t12s1 t21r1 t21s1 t22s1 r s fr(t11) fs(t12) (b) Graph Spreading Fig. 1: Two Schemes for Profile Spreading The weights of the new related terms are proportional to the weights of the original term as the weight wi of a term ti indicates the importance of the term within a user profile. However, during spreading the weights of the related terms should differ accord￾ing to the semantics of the relationships on the edge. For example, spreading based on Wikipedia may be limited to only spreading along the parent categories. We therefore use a set of linear influence functions, one per relationship-type (role/property of an ontology), to control the spreading process. For example, a spreading process based on Wordnet limited to types synonym and antonym can have functions tij = wi × 0.9 and tij = wi × −0.9 respectively. We propose two schemes for representing the related terms post-spreading: extended set and semantic network. 4.1 Set Spreading Depicted in Figure 1a, set spreading is a process of extending an user profile such that the related terms, which are determined with respect to an ontology, are just appended to the original set of terms. Set spreading is an iterative process. After each iteration, the related terms from the previous iterations are appended to the profile. The spreading process is terminated if there are no related terms to spread the profile with or after a fixed number of iterations. 4.2 Graph spreading Shown in Figure 1b, graph spreading is the process where terms from two profiles and the related terms are build into a semantic network (SAN). Unlike set spreading, graph spreading preserves the relationship between a term in a profile and its related term in the form of a graph edge. This allows consideration of relationships based on their semantics on the same network. Graph spreading terminates like set spreading, or if there exists a path between every pair of the term nodes from the two profiles. This condition best suits the ontologies that have a top root element which subsumes the rest of the elements