C Porcel et aL /Expert Systems with Applications 36(2009)5173-518 79 Js冈x=(b0),ifx=100 VR=(b4,0),ifj=100 Ls9x=(bo, 0), otherwise VR=(b3,0),ifj=118 VR1=(bo, 0), otherwise Remark. The UNESCO code 3206 is in the position 100 of the list so it is stored in VU/s[100 Remark. The UNESCO codes 3206 and 3309 are in the position The Fig. 5 shows an example of of a user access 00 and 118 of the list so they are stored respectively in VRi[100 and vR[118 3.3. 2 Resources insertion process An example of resource list is shown in Fig. 6 This sub-process is carried out by the experts, i.e. the transf chnology technicians that receive or find information about a re- 3.3.3. Filtering process source and they want to spread this information. The experts insert As we have said, we use the vector model(Korfhage, 1997)to the interesting resources into the system and it automatically represent the resource scope and the user's sends the information to the suitable users along with a relevance vector model uses similarity calculations to do the matching pro- degree and collaborations possibilities. cess, such as Euclidean Distance or Cosine Measure. Exactly we As we said in the previous section, the system stores the general use the standard cosine measure(Korfhage, 1997). However, as information about the resource and its scope. The scope is repre sented by a vector of UNESCO codes whereby to insert the resource similarity measure we have linguistic values, we need to introduce a new linguistic the experts decide the UNESCo codes to assign it. Moreover, to manage the linguistic information, the experts also decide a lin- guistic 2-tuple (bi, 24), with b, E S1, to weight the importance de- 01(VR, VU)=4 Ck(4(rk, ark)x4(uk, auk)) gree of each UNESCO code of level 2 with regard to the resource V∑k=1(4(rk,x)2xV∑k=(A(uk,) Hence. to the system, insert all the information about it and number of UNESCO codes of level 2),(k, ark)is the 2-tuple linguistic en the experts are going to insert a new resource, where n is the number of terms used to define the vectors(i.e.the th nally they assess the importance degree of each UNESCO cod value of term k in the resource vector (VR)(uk, auk)is its 2-tuple lin- of level 2 with regard to the resource. To do this, the system shows guistic value in the user vector(VU). With this similarity measure a list of UNESCO codes of level 2 and the experts decide the codes we obtain a linguistic value to assess the similarity between a re- to assign to the resource scope, selecting a code of the list and source and a user. In the case of two users or two resources. this lin- signing it a linguistic label to assess its importance degree. Then guistic similarity measure can be applied in a similar way they accept and can either add another UNESco code or finally the Following this approach, when a new resource has been il resource insertion serted into the system, we compute the linguistic similarity mea sure a(VR, vU)) between the new resource scope vector (VR) Example 2. Now let us suppose the expert receives a call i about a against all the user vectors(VU j=l,., m where m is the num- e research resource. Then, he/she inserts the call ber of users of the system)to find the fit users to deliver this infor into the system, introducing all the available information and mation. If or(VR, VU/)>y, the user is chosen. Previously we have selecting from a list the UNESCO codes which match with the call. defined a linguistic threshold value(v)to filter out the informa- In this example, the expert could select the codes 3206-Science of tion. In this iteration, the system takes into account also the user Nutriment with importance degree Total(b4, 0), witha E S1) and preferences(kind of resources and amounts)to consider the user 3309-Food Technology with degree Very High((b3, 0), withbg E S1). or not. The collaboration preferences are used to classify the se- Once the expert inserts this information, we have a vector VR lected users in two sets, collaborators c and non-collaborators defining the resource i with the following values SIREZIN Recommender System about Research Resources urie Builaing. Campus o 14071 Cordoba Fig. 5. Example of a user access.VUID½x¼ðb4; 0Þ; if x ¼ 100 VUID½x¼ðb0; 0Þ; otherwise: Remark. The UNESCO code 3206 is in the position 100 of the list so it is stored in VUID½100. The Fig. 5 shows an example of a user access. 3.3.2. Resources insertion process This sub-process is carried out by the experts, i.e., the transfer technology technicians that receive or find information about a re￾source and they want to spread this information. The experts insert the interesting resources into the system and it automatically sends the information to the suitable users along with a relevance degree and collaborations possibilities. As we said in the previous section, the system stores the general information about the resource and its scope. The scope is repre￾sented by a vector of UNESCO codes whereby to insert the resource the experts decide the UNESCO codes to assign it. Moreover, to manage the linguistic information, the experts also decide a lin￾guistic 2-tuple ðbi; aiÞ, with bi 2 S1, to weight the importance de￾gree of each UNESCO code of level 2 with regard to the resource scope. Hence, when the experts are going to insert a new resource, they access to the system, insert all the information about it and finally they assess the importance degree of each UNESCO code of level 2 with regard to the resource. To do this, the system shows a list of UNESCO codes of level 2 and the experts decide the codes to assign to the resource scope, selecting a code of the list and assigning it a linguistic label to assess its importance degree. Then they accept and can either add another UNESCO code or finally the resource insertion. Example 2. Now let us suppose the expert receives a call i about a nutriment science research resource. Then, he/she inserts the call into the system, introducing all the available information and selecting from a list the UNESCO codes which match with the call. In this example, the expert could select the codes 3206 – Science of Nutriment with importance degree Total (ðb4; 0Þ; withb4 2 S1) and 3309 – Food Technology with degree Very High (ðb3; 0Þ; withb3 2 S1). Once the expert inserts this information, we have a vector VRi defining the resource i with the following values: VRi½j¼ðb4; 0Þ; if j ¼ 100 VRi½j¼ðb3; 0Þ; if j ¼ 118 VRi½j¼ðb0; 0Þ; otherwise: Remark. The UNESCO codes 3206 and 3309 are in the positions 100 and 118 of the list so they are stored respectively in VRi½100 and VRi½118. An example of resource list is shown in Fig. 6. 3.3.3. Filtering process As we have said, we use the vector model (Korfhage, 1997) to represent the resource scope and the user’s topics of interest. This vector model uses similarity calculations to do the matching pro￾cess, such as Euclidean Distance or Cosine Measure. Exactly we use the standard cosine measure (Korfhage, 1997). However, as we have linguistic values, we need to introduce a new linguistic similarity measure: rlðVR; VUÞ ¼ D Pn k¼1ðD1 ðrk; arkÞ  D1 ðuk; aukÞÞ ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi Pn k¼1ðD1 ðrk; arkÞÞ2 q  ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi Pn k¼1ðD1 ðuk; aukÞÞ2 q 0 B@ 1 CA; where n is the number of terms used to define the vectors (i.e. the number of UNESCO codes of level 2), ðrk; arkÞ is the 2-tuple linguistic value of term k in the resource vector (VR), ðuk; aukÞ is its 2-tuple lin￾guistic value in the user vector (VU). With this similarity measure we obtain a linguistic value to assess the similarity between a re￾source and a user. In the case of two users or two resources, this lin￾guistic similarity measure can be applied in a similar way. Following this approach, when a new resource has been in￾serted into the system, we compute the linguistic similarity mea￾sure rlðVRi; VUjÞ between the new resource scope vector (VRi) against all the user vectors (VUj, j ¼ 1; ... ; m where m is the num￾ber of users of the system) to find the fit users to deliver this infor￾mation. If rlðVRi; VUjÞ P w, the user j is chosen. Previously we have defined a linguistic threshold value (w) to filter out the informa￾tion. In this iteration, the system takes into account also the user preferences (kind of resources and amounts) to consider the user or not. The collaboration preferences are used to classify the se￾lected users in two sets, collaborators UC and non-collaborators UN. Fig. 5. Example of a user access. C. Porcel et al. / Expert Systems with Applications 36 (2009) 5173–5183 5179