Knowledge-Based Systems 23(2010)32 Contents lists available at Science Direct Knowledge-Based Systems ELSEVIER journalhomepagewww.elsevier.com/locate/knosys Dealing with incomplete information in a fuzzy linguistic recommender system to disseminate information in university digital libraries C Porcel , E. Herrera-Viedma b University of Granada, Dep of Computer Science and Artificial Intelligence, Granada, Spain ARTICLE IN FO A BSTRACT As in the Web, the growing of information is the main problem of the academic digital libraries. Thus, ible online 5 August 2009 milar tools could be applied in university digital libraries to facilitate the information access by the stu- dents and teachers. In [46 we presented a fuzzy linguistic recommender system to advice research esources in university digital libraries. The problem of this system is that the user profiles are provide recommender systen directly by the own users and the process for acquiring user preferences is quite difficult because it niversity digital libraries ncomplete fuzzy linguistic preference facilitates the acquisition of the user preferences to characterize the user profiles. We allow users to provide their preferences by means of incomplete fuzzy linguistic preference relation. We include tools o manage incomplete information when the users express their preferences, and, in such a way, we show that the acquisition of the user profiles is improved. e 2009 Elsevier B V. All rights reserved. 1 Introduction Recommender systems are becoming popular tools for reducing information overload and to improve the sales in e-commerce web Digital libraries are information collections that have associated sites [ 7, 9, 35, 40, 49. The use of this kind of systems allows to re services delivered to user communities using a variety of technol- ommend resources interesting for the users, at the same time that ogies[8, 15.48. Therefore, digital libraries are the logical exten- these resources are inserted into the system. In the UDL frame sions of physical libraries in the electronic information society. work, recommender systems [7, 49] can be used to help users These extensions amplify existing resources and services. As such, (teachers, students and library staff) to find out and select their digital libraries offer new levels of access to broader audiences of information and knowledge sources[43 sers and new opportunities for the library In practice, a digital li- Generally, in a recommender system the users'information brary makes its contents and services remotely accessible through preferences can be used to define user profiles that are applied networks such as the Web or limited-access intranets 39, 50]. as filters to streams of documents[7, 47,49 In 45, 46 we devel- As digital libraries become commonplace and as their contents oped some recommender systems in an academic context. For in- become more varied, the users expect more sophisticated services stance in [45 we proposed a fuzzy linguistic recommender om them[8, 15,48, 50 A service that is particularly important is system for a technology transfer office which helps researchers the selective dissemination of information or filtering, to help the and environment companies allowing them to obtain information users to access interesting information for them. Users develop automatically about research resources(calls or projects)in their interest profiles and as new materials(books, papers, reports, interest areas; in [46 we proposed a fuzzy linguistic recommender and so on)are added to the collection, they are compared to the system to achieve major advances in the activities of UDl, which profiles and relevant items are sent to the users [39 recommends researchers specialize Moreover, digital libraries have been applied in a lot of contexts tary resources related with their respective research areas. The ut in this paper we focus on an academic environment. University problem of both recommender systems is that users must directly Digital Libraries(UDL) provide information resources and services specify their user profiles by providing their preferences on all top to students, faculty and staff in an environment that supports ics of interest and it requires too much user effort. learning, teaching and research [11 In th s paper, we focus on the idea of that tem could be seen as a decision support system(DSS)[37, 38, 44 g author. where the solution alternatives are the digital resources inserted -mail cporceleujaenes (C. Porcel) viedmaedecsai ugres into the library, and the criteria to satisfy are the user profiles The proper use of these recommendation systems is essential to 0950-7051/s-see front matter a 2009 Elsevier B v. All rights reserved oi:10.1016 knosys.200907.007
Dealing with incomplete information in a fuzzy linguistic recommender system to disseminate information in university digital libraries C. Porcel a,*, E. Herrera-Viedma b aUniversity of Jaén, Department of Computer Science, Jaén, Spain bUniversity of Granada, Department of Computer Science and Artificial Intelligence, Granada, Spain article info Article history: Available online 5 August 2009 Keywords: Recommender systems Fuzzy linguistic modeling University digital libraries Incomplete fuzzy linguistic preference relation abstract As in the Web, the growing of information is the main problem of the academic digital libraries. Thus, similar tools could be applied in university digital libraries to facilitate the information access by the students and teachers. In [46] we presented a fuzzy linguistic recommender system to advice research resources in university digital libraries. The problem of this system is that the user profiles are provided directly by the own users and the process for acquiring user preferences is quite difficult because it requires too much user effort. In this paper we present a new fuzzy linguistic recommender system that facilitates the acquisition of the user preferences to characterize the user profiles. We allow users to provide their preferences by means of incomplete fuzzy linguistic preference relation. We include tools to manage incomplete information when the users express their preferences, and, in such a way, we show that the acquisition of the user profiles is improved. 2009 Elsevier B.V. All rights reserved. 1. Introduction Digital libraries are information collections that have associated services delivered to user communities using a variety of technologies [8,15,48]. Therefore, digital libraries are the logical extensions of physical libraries in the electronic information society. These extensions amplify existing resources and services. As such, digital libraries offer new levels of access to broader audiences of users and new opportunities for the library. In practice, a digital library makes its contents and services remotely accessible through networks such as the Web or limited-access intranets [39,50]. As digital libraries become commonplace and as their contents become more varied, the users expect more sophisticated services from them [8,15,48,50]. A service that is particularly important is the selective dissemination of information or filtering, to help the users to access interesting information for them. Users develop interest profiles and as new materials (books, papers, reports, and so on) are added to the collection, they are compared to the profiles and relevant items are sent to the users [39]. Moreover, digital libraries have been applied in a lot of contexts but in this paper we focus on an academic environment. University Digital Libraries (UDL) provide information resources and services to students, faculty and staff in an environment that supports learning, teaching and research [11]. Recommender systems are becoming popular tools for reducing information overload and to improve the sales in e-commerce web sites [7,9,35,40,49]. The use of this kind of systems allows to recommend resources interesting for the users, at the same time that these resources are inserted into the system. In the UDL framework, recommender systems [7,49] can be used to help users (teachers, students and library staff) to find out and select their information and knowledge sources [43]. Generally, in a recommender system the users’ information preferences can be used to define user profiles that are applied as filters to streams of documents [7,47,49]. In [45,46] we developed some recommender systems in an academic context. For instance, in [45] we proposed a fuzzy linguistic recommender system for a technology transfer office which helps researchers and environment companies allowing them to obtain information automatically about research resources (calls or projects) in their interest areas; in [46] we proposed a fuzzy linguistic recommender system to achieve major advances in the activities of UDL, which recommends researchers specialized resources and complementary resources related with their respective research areas. The problem of both recommender systems is that users must directly specify their user profiles by providing their preferences on all topics of interest and it requires too much user effort. In this paper, we focus on the idea of that a recommender system could be seen as a decision support system (DSS) [37,38,44], where the solution alternatives are the digital resources inserted into the library, and the criteria to satisfy are the user profiles. The proper use of these recommendation systems is essential to 0950-7051/$ - see front matter 2009 Elsevier B.V. All rights reserved. doi:10.1016/j.knosys.2009.07.007 * Corresponding author. E-mail addresses: cporcel@ujaen.es (C. Porcel), viedma@decsai.ugr.es (E. Herrera-Viedma). Knowledge-Based Systems 23 (2010) 32–39 Contents lists available at ScienceDirect Knowledge-Based Systems journal homepage: www.elsevier.com/locate/knosys
provide real personalized services, and it can substantially reduce A variety of techniques have been proposed as the basis for I formation overload and increase user satisfaction. Therefore, it ommender systems [7, 17, 40, 49: all of these techniques have ben- has become an important area in information systems and decision efits and disadvantages. The use of an hybrid approach is proposed upport research[37, 38, 44 So, the activity of a recommender sys- to smooth out the disadvantages of each one of them and to exploit tem can be seen as a group decision making(GDM)problem, so we their benefits [5, 13, 16]. In these kind of systems, the users'infor- can adopt the typical representation formats used in GDM, as for mation preferences can be used to define user profiles that are ap- example, fuzzy preference relations [19, 20, 28, 32, 41 This repre- plied as filters to streams of documents. The construction of sentation format presents a high expressivity and some interesting accurate profiles is a key task and the system s success will depend properties that allow us to work easily. However, in real world on a large extent on the ability of the learned profiles to represent problems it is common to find situations in which users are not the user preferences [47] able to provide all the preference values that are required, and The recommendation activity is followed by a relevance hen, we have to deal with incomplete fuzzy preference relations back phase. Relevance feedback is a cyclic process whereby the 1-3.2526,41 users feed back into the system decisions on the relevance of re- The aim of this paper is to present a new fuzzy linguistic recom- trieved documents and the system uses these evaluations to auto- mender defined in a UDL framework which overcomes the problem matically update the user profiles [17, 49. of user profile characterization observed in the recommender sys tems defined in[45, 46]. In order to improve the system perfor- 2.2. The 2-tuple fuzzy linguistic approach mance, we propose an alternative way to obtain accurate and useful knowledge about the user preferences. This new recom- The fuzzy linguistic modeling(FLM)is a tool based on the con- mender system allows users to provide their preferences by means cept of linguistic variable[52] which has given very good results for of incomplete fuzzy linguistic preference relations [1. and in such modeling qualitative information in many problems, e.g., in deci- a way, we facilitate users the expression of their preferences and, sion making 20], quality evaluation 33, 34, models of information consequently, the determination of user profiles process. The rec retrieval 23, 24, 29-31, political analysis [4. etc. ommender system is able to complete the incomplete preference The 2-tuple FLM [21 is a continuous model of representation relations using the tools proposed in [1, 2. 26. Each user profile is of information that allows to reduce the loss of information typi composed of both user preferences on topics of interest and user cal of other fuzzy linguistic approaches(classical and ordinal preferences on collaboration possibilities with other users. Then, [18,)). the recommender system is able to recommend both research re Lets=iso,., Sg be a linguistic term set with odd cardinali sources and collaboration possibilities to the users of a UDL. As where the mid term represents a indifference value and the rest of in[45, 46] we define this recommender system in a multi-granular the terms are symmetrically related to it. We assume that the fuzzy linguistic context [10, 12, 22, 27, 32, 42 In such a way, we semantics of labels is given by means of triangular membership incorporate in the recommender system flexible tools to handle functions and consider all terms distributed on a scale on which the information by allowing to represent the different concepts a total order is defined, S <S,<i<j. In this fuzzy linguisti of the system with different linguistic label sets. context, if a symbolic method [18, 20 aggregating linguistic infor- The rest of the paper is set out as follows. Section 2 presents the mation obtains a value BE[0, g], and B#10,.,g, then an preliminaries necessary to develop the proposed model. Section 3 approximation function is used to express the result in S B is rep- presents the new recommender system to the dissemination of resented by means of 2-tuples (Si, a4), S E Sand a E[-5,5)where knowledge in a UDL Section 4 reports the system evaluation and 5, represents the linguistic label of the information, and o, is a the experimental results. Finally, our conclusions are pointed out numerical value expressing the value of the translation from the in Section 5 original result B to the closest index label, i, in the linguistic term set(S S). This 2-tuple representation model defines a set of transformation functions between numeric values and 2-tuple △(B)=(S1,x)and△-l(s,x)=B∈0g][21 The computational model is defined by presenting a negation 2.1. Recommender systems operator, comparison of 2-tuples and aggregation operators [21]. Using functions A and a"that transform without loss of informa- Recommender systems could be defined as systems that pre tion numerical values into linguistic 2-tuples and viceversa, any of duce individualized recommendations as output, or have the effect the existing aggregation operators can be easily extended for of guiding the user in a personalized way to interesting or useful dealing with linguistic 2-tuples. Some examples are objects in a large space of possible options [6]. It is a research area that offers tools for discriminating between relevant and irrelevant information by providing personalized assistance for continuous information accesses [43, 49 Automatic filtering services differ from retrieval services 23, 24, 29-31]in that in filtering the corpus changes continuously, the users have long time information needs(described by means of user profiles)in- stead of introducing a query into the system, and their objective is to remove irrelevant data from incoming streams of data items Definition 2(Weighted an Let x=i(r, a1) [17, 39, 49]. A result from a recommender system is understood as (rn, an)i be a set of linguistic d W=wI hts. the a recommendation, an option worthy of consideration, while a re- ghted average Xw is sult from an information retrieval system is interpreted as a match to the users query 7]. However both systems present some anal ogies, and in this sense they could be considered a DSS [44]. In both x"[(1, a1),.,(rm,an)]=A 14(r,a)W cases. the solution alternatives would be the documents to recom- mend or retrieve and the criteria to satisfy would be the user pro- files and user queries, respectively
provide real personalized services, and it can substantially reduce information overload and increase user satisfaction. Therefore, it has become an important area in information systems and decision support research [37,38,44]. So, the activity of a recommender system can be seen as a group decision making (GDM) problem, so we can adopt the typical representation formats used in GDM, as for example, fuzzy preference relations [19,20,28,32,41]. This representation format presents a high expressivity and some interesting properties that allow us to work easily. However, in real world problems it is common to find situations in which users are not able to provide all the preference values that are required, and then, we have to deal with incomplete fuzzy preference relations [1–3,25,26,41]. The aim of this paper is to present a new fuzzy linguistic recommender defined in a UDL framework which overcomes the problem of user profile characterization observed in the recommender systems defined in [45,46]. In order to improve the system performance, we propose an alternative way to obtain accurate and useful knowledge about the user preferences. This new recommender system allows users to provide their preferences by means of incomplete fuzzy linguistic preference relations [1], and in such a way, we facilitate users the expression of their preferences and, consequently, the determination of user profiles process. The recommender system is able to complete the incomplete preference relations using the tools proposed in [1,2,26]. Each user profile is composed of both user preferences on topics of interest and user preferences on collaboration possibilities with other users. Then, the recommender system is able to recommend both research resources and collaboration possibilities to the users of a UDL. As in [45,46] we define this recommender system in a multi-granular fuzzy linguistic context [10,12,22,27,32,42]. In such a way, we incorporate in the recommender system flexible tools to handle the information by allowing to represent the different concepts of the system with different linguistic label sets. The rest of the paper is set out as follows. Section 2 presents the preliminaries necessary to develop the proposed model. Section 3 presents the new recommender system to the dissemination of knowledge in a UDL. Section 4 reports the system evaluation and the experimental results. Finally, our conclusions are pointed out in Section 5. 2. Preliminaries 2.1. Recommender systems Recommender systems could be defined as systems that produce individualized recommendations as output, or have the effect of guiding the user in a personalized way to interesting or useful objects in a large space of possible options [6]. It is a research area that offers tools for discriminating between relevant and irrelevant information by providing personalized assistance for continuous information accesses [43,49]. Automatic filtering services differ from retrieval services [23,24,29–31] in that in filtering the corpus changes continuously, the users have long time information needs (described by means of user profiles) instead of introducing a query into the system, and their objective is to remove irrelevant data from incoming streams of data items [17,39,49]. A result from a recommender system is understood as a recommendation, an option worthy of consideration, while a result from an information retrieval system is interpreted as a match to the user’s query [7]. However both systems present some analogies, and in this sense they could be considered a DSS [44]. In both cases, the solution alternatives would be the documents to recommend or retrieve and the criteria to satisfy would be the user pro- files and user queries, respectively. A variety of techniques have been proposed as the basis for recommender systems [7,17,40,49]; all of these techniques have benefits and disadvantages. The use of an hybrid approach is proposed to smooth out the disadvantages of each one of them and to exploit their benefits [5,13,16]. In these kind of systems, the users’ information preferences can be used to define user profiles that are applied as filters to streams of documents. The construction of accurate profiles is a key task and the system’s success will depend on a large extent on the ability of the learned profiles to represent the user preferences [47]. The recommendation activity is followed by a relevance feedback phase. Relevance feedback is a cyclic process whereby the users feed back into the system decisions on the relevance of retrieved documents and the system uses these evaluations to automatically update the user profiles [17,49]. 2.2. The 2-tuple fuzzy linguistic approach The fuzzy linguistic modeling (FLM) is a tool based on the concept of linguistic variable[52] which has given very good results for modeling qualitative information in many problems, e.g., in decision making [20], quality evaluation [33,34], models of information retrieval [23,24,29–31], political analysis [4], etc. The 2-tuple FLM [21] is a continuous model of representation of information that allows to reduce the loss of information typical of other fuzzy linguistic approaches (classical and ordinal [18,52]). Let S ¼ fs0; ... ; sg g be a linguistic term set with odd cardinality, where the mid term represents a indifference value and the rest of the terms are symmetrically related to it. We assume that the semantics of labels is given by means of triangular membership functions and consider all terms distributed on a scale on which a total order is defined, si 6 sj () i 6 j. In this fuzzy linguistic context, if a symbolic method [18,20] aggregating linguistic information obtains a value b 2 ½0; g, and b R f0; ... ; gg; then an approximation function is used to express the result in S. b is represented by means of 2-tuples ðsi; aiÞ, si 2 S and ai 2 ½:5; :5Þ where si represents the linguistic label of the information, and ai is a numerical value expressing the value of the translation from the original result b to the closest index label, i, in the linguistic term set ðsi 2 SÞ. This 2-tuple representation model defines a set of transformation functions between numeric values and 2-tuples DðbÞ¼ðsi; aÞ and D1 ðsi; aÞ ¼ b 2 ½0; g [21]. The computational model is defined by presenting a negation operator, comparison of 2-tuples and aggregation operators [21]. Using functions D and D1 that transform without loss of information numerical values into linguistic 2-tuples and viceversa, any of the existing aggregation operators can be easily extended for dealing with linguistic 2-tuples. Some examples are Definition 1 (Arithmetic mean). Let x ¼ fðr1; a1Þ; ... ;ðrn; anÞg be a set of linguistic 2-tuples, the 2-tuple arithmetic mean xe is computed as, xe ½ ¼ ð Þ r1; a1 ; ... ; ð Þ rn; an D Xn i¼1 1 n D1 ri ð Þ ; ai ! ¼ D 1 n Xn i¼1 bi !: ð1Þ Definition 2 (Weighted average operator). Let x ¼ fðr1; a1Þ; ... ; ðrn; anÞg be a set of linguistic 2-tuples and W ¼ fw1; ... ; wng be their associated weights. The 2-tuple weighted average xw is xw½ ¼ ð Þ r1; a1 ; ... ; ð Þ rn; an D Pn i¼1D1 ri P ð Þ ; ai wi n i¼1wi ! ¼ D Pn i¼1bi P wi n i¼1wi : ð2Þ C. Porcel, E. Herrera-Viedma / Knowledge-Based Systems 23 (2010) 32–39 33
C Porcel, E. Herrera-Viedma/ Knowledge-Based Systems 23(2010)32-39 TF:t,n(t)→lt,n() Linguistic hierarchies. s n(D), ar())-(n(t2)-1 Level- TF(st, e)=n(o) l(2.13) As it was pointed out in[22 this family of transformation fund tions is bijective. This result guarantees that the transformations between levels of a linguistic hierarchy are carried out without loss 2.3. Incomplete fuzzy preference relations Definition 5. A fuzzy preference relation P on a set of alternatives X=x1, ...,xn) is a fuzzy set on the product set X xX, i.e., it is characterized by a mem nction Ap:XxX→0,1] When cardinality of X is small, the preference relation may be conveniently represented by the n x n matrix P=(Pig), being Py=Ap(xi, xy)(Vi, jE1,.,nh)interpreted as the preference degree or intensity of the alternative x over x, where . P= 1/2 indicates indifference between x and Py=1 indicates that x is absolutely preferred to x. and Py>1/2 indicates that x is preferred to x. Fig. 1. Linguistic hierarchy of 3, 5 and 9 labels. However, as we have mentioned, our syster rates the multi-granular FLM based on 2-tuples, so we must define a linguis- tic preference relation as follows: Definition 6. Let X=x1.,xn) a set of alternatives and sa Let x=i(r1, linguistic term set. A linguistic preference relation P= Pii, je weights. The pp:X×X→S×10.5.0.5), where Py=Ap(, x)is a 2-tuple which denotes the preference de- 对(,2,(wx)…(m)、%2=4/2B- gree of alternative x, regarding to x. As aforementioned, in many real world GDM problems the ex- (3) perts are often not able to provide all the preference values that are required In order to model these situations, we use incomplete with B,=A (n, a,)and Bw=A"(Wi, a) fuzzy preference relations [1-3, 25, 26, 41] nportant parameter to Definition 7. A function f: X-Y is partial when not every determine is the"granularity of uncertainty". ie, the cardinality element in the set x necessarily maps onto an element in the set of the linguistic term set S. when different experts have different Y. when every element from the set X maps onto one element of uncertainty degrees on the phenomenon or when an expert has the set Y, then we have a total function. to assess different concepts, then several linguistic term sets wit a different granularity of uncertainty are necessary [22, 32. In(22) Definition 8. A two-tuple fuzzy linguistic preference relation Pon a multi-granular 2-tuple FLM based on the concept of linguistic a set of alternatives X with a partial membership function is an hierarchy is proposed. incomplete two-tuple fuzzy linguistic preference relation A linguistic hierarchy(LH), is a set of levels l(t, n(t)), where each level t is a linguistic term set with different granularity n(t) from 3. A recommender system for the dissemination of information according to their granularity, i.e. a level t +1 provides a linguistic in UDL using incomplete linguistic preference relations refinement of the previous level t. We can define a level from its redecessor level as I(t, n(t))-I(t+1, 2. n(t)-1) Table 1 shows The UDL staff manage and spread many information about re- le granularity needed in each linguistic term set of the level t search resources such as electronic books, electronic papers, depending on the value n(t) defined in the first level (3 and 7. onic journals, official dailies and so on [8, 48. Nowadays, this amount of information is growing up and they are in need of auto- A graphical example of a linguistic hierarchy is shown in Fig. 1 mated tools to filter and spread that information to the users in a In [22] a family of transformation functions between labels simple and timely manner. On the other hand, UDL need from different levels was introduced tools to help them to insert their preferences to form accurate ofi Definition 4. Let LH=U(t, n(t)) be a linguistic hierarchy whose In this section we present a new fuzzy linguistic recommender i-1. The system in which the user transformation function between a 2-tuple that belongs to level t ences represented by inco fuzzy linguistic preference rela- and another 2-tuple in level t/t is defined as tions [1. This proposal es with some advantages with
Definition 3 (Linguistic weighted average operator). Let x ¼ fðr1; a1Þ; ... ;ðrn; anÞg be a set of linguistic 2-tuples and W ¼ fðw1; aw 1 Þ; ... ; ðwn; aw n Þg be their linguistic 2-tuple associated weights. The 2-tuple linguistic weighted average xw l is xw l ð Þ r1; a1 ; w1; aw 1 ... ð Þ rn; an ; wn; aw n ¼ D Pn i¼1bi P bWi n i¼1bWi !; ð3Þ with bi ¼ D1 ðri; aiÞ and bWi ¼ D1 ðwi; aw i Þ. In any fuzzy linguistic approach, an important parameter to determine is the ‘‘granularity of uncertainty”, i.e., the cardinality of the linguistic term set S. When different experts have different uncertainty degrees on the phenomenon or when an expert has to assess different concepts, then several linguistic term sets with a different granularity of uncertainty are necessary [22,32]. In [22] a multi-granular 2-tuple FLM based on the concept of linguistic hierarchy is proposed. A linguistic hierarchy (LH), is a set of levels lðt; nðtÞÞ, where each level t is a linguistic term set with different granularity nðtÞ from the remaining of levels of the hierarchy. The levels are ordered according to their granularity, i.e., a level t þ 1 provides a linguistic refinement of the previous level t. We can define a level from its predecessor level as lðt; nðtÞÞ ! lðt þ 1; 2 nðtÞ 1Þ. Table 1 shows the granularity needed in each linguistic term set of the level t depending on the value n(t) defined in the first level (3 and 7, respectively). A graphical example of a linguistic hierarchy is shown in Fig. 1. In [22] a family of transformation functions between labels from different levels was introduced: Definition 4. Let LH ¼ S tlðt; nðtÞÞ be a linguistic hierarchy whose linguistic term sets are denoted as SnðtÞ ¼ fs nðtÞ 0 ; ... ; s nðtÞ nðtÞ1g. The transformation function between a 2-tuple that belongs to level t and another 2-tuple in level t0–t is defined as TFt t0 : l tð Þ! ; nðtÞ l t0 ; nðt 0 ð ÞÞ ; TFt t0 s nðtÞ i ; anðtÞ ¼ D D1 s nðtÞ i ; anðtÞ nðt 0 ð Þ Þ 1 nðtÞ 1 0 @ 1 A: As it was pointed out in [22] this family of transformation functions is bijective. This result guarantees that the transformations between levels of a linguistic hierarchy are carried out without loss of information. 2.3. Incomplete fuzzy preference relations Definition 5. A fuzzy preference relation P on a set of alternatives X ¼ fx1; ... ; xng is a fuzzy set on the product set X X, i.e., it is characterized by a membership function lP : X X ! ½0; 1: When cardinality of X is small, the preference relation may be conveniently represented by the n n matrix P ¼ ðpijÞ, being pij ¼ lPðxi; xjÞ ð8i; j 2 f1; ... ; ngÞ interpreted as the preference degree or intensity of the alternative xi over xj, where pij ¼ 1=2 indicates indifference between xi and xj, pij ¼ 1 indicates that xi is absolutely preferred to xj, and pij > 1=2 indicates that xi is preferred to xj. However, as we have mentioned, our system integrates the multi-granular FLM based on 2-tuples, so we must define a linguistic preference relation as follows: Definition 6. Let X ¼ fx1 ... ; xng a set of alternatives and S a linguistic term set. A linguistic preference relation P ¼ pijð8i; j 2 f1; ... ; ngÞ on X is lP : X X ! S ½ Þ 0:5; 0:5 ; ð4Þ where pij ¼ lPðxi; xjÞ is a 2-tuple which denotes the preference degree of alternative xi regarding to xj. As aforementioned, in many real world GDM problems the experts are often not able to provide all the preference values that are required. In order to model these situations, we use incomplete fuzzy preference relations [1–3,25,26,41]. Definition 7. A function f : X ! Y is partial when not every element in the set X necessarily maps onto an element in the set Y. When every element from the set X maps onto one element of the set Y, then we have a total function. Definition 8. A two-tuple fuzzy linguistic preference relation P on a set of alternatives X with a partial membership function is an incomplete two-tuple fuzzy linguistic preference relation. 3. A recommender system for the dissemination of information in UDL using incomplete linguistic preference relations The UDL staff manage and spread many information about research resources such as electronic books, electronic papers, electronic journals, official dailies and so on [8,48]. Nowadays, this amount of information is growing up and they are in need of automated tools to filter and spread that information to the users in a simple and timely manner. On the other hand, UDL users need tools to help them to insert their preferences to form accurate profiles. In this section we present a new fuzzy linguistic recommender system in which the user profiles are obtained from user preferences represented by incomplete fuzzy linguistic preference relations [1]. This proposal contributes with some advantages with Table 1 Linguistic hierarchies. Level 1 Level 2 Level 3 l(t,n(t)) l(1,3) l(2,5) l(3,9) l(t,n(t)) l(1,7) l(2,13) Fig. 1. Linguistic hierarchy of 3, 5 and 9 labels. 34 C. Porcel, E. Herrera-Viedma / Knowledge-Based Systems 23 (2010) 32–39
C Porcel, E. Herrera-Viedma/ Knowledge-Based Systems 23(2010)32-39 Resources presentations 以)<~uo-cy} Recom Fig. 2. Operating scheme. regard to previous systems [46, 45 because it facilitates the 3. 1. Resources representation expression of their preferences to the users and reduces the user The considered resources are journal articles, conference contri- effort to characterize their user profiles. It is applied ise butions, book chapters, books or edited books. Once the library UDL users on the best research resources that could satisfy their staff insert all the available information about a new resource information needs in UDL Moreover, the system recommends col- the system obtains an internal representation mainly based in laboration possibilities to meet another researchers of related the resource scope. We use the vector model [36] to represent the areas which could collaborate with them in projects or interest resource scope. Thus, to sent a resource i we use a classifica- works. In such a way, this new recommender system improves tion composed by 25 disciplines (see Fig 3). In each position we the services that a UDL provides to the users, because it is easier store a linguistic 2-tuple value representing the importance degree the time cost to establish the user profile d it allows to decrease of the resource scope with respect to the discipline represented by to obtain the knowledge about the users an that position In Fig. 2 we can see the basic operating scheme, which is ex- plained in the following subsections WR=(WR1,WR2,…,VR25). Then, each component VRy E SI, withj=(1,., 25]. indicates the linguistic importance degree of the discipline j with regard to In the proposed system, the user-system communication is car- the resource i. These importance degrees are assigned by the ed out by using a multi-granular fuzzy linguistic approach library staff when they add a new resource. 22, 32, in order to allow a higher flexibility in the communication processes of the system. The system uses different label sets (S1, S2,.)to represent the different concepts to be assessed in 3. 1.2. User profiles its filtering activity. These label sets, Si, are chosen from those label The user profiles are composed of two kinds of user prefe sets that compose a lh, i. e, S E LH. We should point out that the number of different label sets that we can use is limited by the (1)User preferences on topics of interest, and umber of levels of LH, and therefore, in many cases the label set (2)User preferences on collaboration possibility with other Si and si can be associated to a same label set of LH but with diffe users take into account the follow g on the concept to be modeled.We lowing concepts that can be assessed in the syste <s The main contribution of this proposal is how users provide eir preferences on topics of interest used to represent the source resources. In previous proposals [45, 46] we represented such user Importance degree of a discipline with respect to a resource preferences using the vector model [36). The problem is that the user preferences(S1). users must insert or edit all the features corresponding to the dis Relevance degree of a resource for a user(S2). ciplines, i. e, in our case 25 categories. Thus, in previous proposal Compatibility degree between two users(S3). we worked with vectors composed of 25 positions(each one corre- Preference degree of a resource regarding another one(S4) sponding to a discipline), but there could exist cases in which this number could be greater. In such a way, users have to perform a To reduce this effort and make the process for ac pics of interest. Following the linguistic hierarchy shown in Fig. 1, in our system great effort to provide their preferences about to we use the level 2(5 labels) to assign importance and egrees(S=S and S4=S), and the level 3(9 labels) eferences easier, in this model we propose an alternative metho levance and compatibility degrees(S2=S and S3 to obtain the user prefe this lh, the linguistic terms in each level are We ask users to provide their preferences on some research re- sources, usually a limited number of resources, four or five. The .S=bo=Null=N, b,=Low=L, b2=Medium =M, b3=High= choice of research resources is made by the personal staff tanking H, b4=Total= TI into account the relevance supplied by the users. As in 41 we pro- S=(co=Null=N, C1= Very_Low=VL, C2= Low=L, C3= pose users to represent their preferences by means of incomplete More_-Low= MLL, C4= Medium= M, Cs== More-Less_ High fuzzy linguistic preference relations. Then, the system presents MLH. C6= High= H, c,=Very -High= VH, cs=Total= TI users only a selection of the most representative resources, and
regard to previous systems [46,45] because it facilitates the expression of their preferences to the users and reduces the user effort to characterize their user profiles. It is applied to advise UDL users on the best research resources that could satisfy their information needs in UDL. Moreover, the system recommends collaboration possibilities to meet another researchers of related areas which could collaborate with them in projects or interest works. In such a way, this new recommender system improves the services that a UDL provides to the users, because it is easier to obtain the knowledge about the users and it allows to decrease the time cost to establish the user profiles. In Fig. 2 we can see the basic operating scheme, which is explained in the following subsections. 3.1. Information representation In the proposed system, the user-system communication is carried out by using a multi-granular fuzzy linguistic approach [22,32], in order to allow a higher flexibility in the communication processes of the system. The system uses different label sets ðS1; S2; ...Þ to represent the different concepts to be assessed in its filtering activity. These label sets, Si, are chosen from those label sets that compose a LH, i.e., Si 2 LH. We should point out that the number of different label sets that we can use is limited by the number of levels of LH, and therefore, in many cases the label sets Si and Sj can be associated to a same label set of LH but with different interpretations, depending on the concept to be modeled. We take into account the following concepts that can be assessed in the system: Importance degree of a discipline with respect to a resource scope or user preferences (S1). Relevance degree of a resource for a user (S2). Compatibility degree between two users (S3). Preference degree of a resource regarding another one (S4). Following the linguistic hierarchy shown in Fig. 1, in our system we use the level 2 (5 labels) to assign importance and preference degrees (S1 ¼ S5 and S4 ¼ S5 ), and the level 3 (9 labels) to assign relevance and compatibility degrees (S2 ¼ S9 and S3 ¼ S9 ). Using this LH, the linguistic terms in each level are S5 ¼ fb0 ¼ Null ¼ N; b1 ¼ Low ¼ L; b2 ¼ Medium ¼ M; b3 ¼ High ¼ H;b4 ¼ Total ¼ Tg S9 ¼ fc0 ¼ Null ¼ N; c1 ¼ Very Low ¼ VL; c2 ¼ Low ¼ L; c3 ¼ More Less Low ¼ MLL; c4 ¼ Medium ¼ M; c5 ¼¼ More Less High MLH; c6 ¼ High ¼ H; c7 ¼ Very High ¼ VH; c8 ¼ Total ¼ Tg 3.1.1. Resources representation The considered resources are journal articles, conference contributions, book chapters, books or edited books. Once the library staff insert all the available information about a new resource, the system obtains an internal representation mainly based in the resource scope. We use the vector model [36] to represent the resource scope. Thus, to represent a resource i, we use a classification composed by 25 disciplines (see Fig. 3). In each position we store a linguistic 2-tuple value representing the importance degree of the resource scope with respect to the discipline represented by that position: VRi ¼ ð Þ VRi1; VRi2; ... ; VRi25 : ð5Þ Then, each component VRij 2 S1, with j ¼ f1; ... ; 25g, indicates the linguistic importance degree of the discipline j with regard to the resource i. These importance degrees are assigned by the library staff when they add a new resource. 3.1.2. User profiles The user profiles are composed of two kinds of user preferences: (1) User preferences on topics of interest, and (2) User preferences on collaboration possibility with other users. The main contribution of this proposal is how users provide their preferences on topics of interest used to represent the source resources. In previous proposals [45,46] we represented such user preferences using the vector model [36]. The problem is that the users must insert or edit all the features corresponding to the disciplines, i.e., in our case 25 categories. Thus, in previous proposals we worked with vectors composed of 25 positions (each one corresponding to a discipline), but there could exist cases in which this number could be greater. In such a way, users have to perform a great effort to provide their preferences about topics of interest. To reduce this effort and make the process for acquiring the user preferences easier, in this model we propose an alternative method to obtain the user preferences on topics of interest. We ask users to provide their preferences on some research resources, usually a limited number of resources, four or five. The choice of research resources is made by the personal staff tanking into account the relevance supplied by the users. As in [41] we propose users to represent their preferences by means of incomplete fuzzy linguistic preference relations. Then, the system presents users only a selection of the most representative resources, and Users Resources Resource representations VRi Acquiring user's preferences Incomplete preference relation P Computing missing information P* Aggregation VUx Recommendations Matching process Fig. 2. Operating scheme. C. Porcel, E. Herrera-Viedma / Knowledge-Based Systems 23 (2010) 32–39 35
C Porcel, E. Herrera-Viedma/ Knowledge-Based Systems 23(2010)32-39 D Agmcuture, animal breeding and faning D vegetal and animal talog anut ecology a Boterhnnlooy, mol d celular biology and genetics v■Fnsi ull L Materials schane e and technotogy a social scienee NA心 mputers stlence and technology 口 Fnurey and comhushbig L Pharmacology and pharmacy dium a Histor and an ns. sonutrustlon and Null 口 medicins and veterinary D Environment and environmental technolo 口 Muttp-discroana NA口spy Null a Psychology and educaton sciences L Chemistny and chemistry tethnoidgr D Telecommunications, electric engineenng, electronics and Fig. 3. Interface to define the disciplines of the resource scope. the users provide their preferences about these resources by mea of an incomplete fuzzy preference relation. Furthermore, according i.e., user preference vector, firstly to results presented in [2. it is enough that the users provide only preference degrees on each consid a row of the preference relation. Then, we use the method pro- to the preference relation P, and posed in [2] to complete the relations. Once the system completes preference degrees together with the vectors that represent ach research resource to obtain the user preference vector. possible to obtain a vector representing the user preferences on The preference degrees coincides with the dominance the topics of interest. Next, we explain this process in detail degrees of a linguistic preference relation [19 To obtain them we propose the application of the arithmetic mean (1)Acquiring the user preferences on a limited number of x(Definition 1). Then, the preference degree of the research resources: At the beginning, the main goal is to resource i for the expert called DGi, is computed as follows: help the users to provide their preferences assuring that these preferences are as consistent as possible. The system DG=X Pi.,Ps] shows users the five most representative resources, R (r1., Ts), and asks them to express their preferences by VUx, .. VUxs) from the aggregation of the vectors that represents neans of an incomplete fuzzy linguistic preference relation the characteristics of the chosen research resources, i.e. ( VRI (see Fig. 4). The users only fill those preferences that they VRs weighted by means of the user preference degrees wish, assigning labels of S4. In the preference relation, each (DGI ... DGs). To do that, we use the linguistic weighted average preference value Py represents the linguistic preference operator defined in Definition 3, and then each position degree of resource i over the resource j according to the user k=(1,., 25)of the vector VUx, is computed as follows provide a relation with only one row of preference values: VUxk=X [(VRik, DG1),., (VRsk, DG5) P12P13P14P15 On the other hand, to complete the user profile, the system asks XXX every user to express his/her collaboration preferences, i. e, if he/ XX (6 ons on collaboration possibili- ties with others users. This could help users to develop multi-dis- ciplinar works or participate in collaborative research projects [46]. They should respond to this question with"Yes"or"No Then, the system completes the preference relation P using the od proposed in [2 and obtains the relation P- 3. 2. Recommendation strategy P12P13 In this phase the system generates the recommendations to de- P= P31 P32 P34 P3s (7) liver the information resources to the fitting users. This process is P41P42P43 based on a matching process developed between user profiles and P51 P52 P53 P54 resource representations [17, 36. To do that, we can use different kinds of similarity measures, such as euclidean distance or Cosine Measure. Particularly we use the standard cosine measure[36].As where Py E Sa are the degrees inserted by the user about the pref- the components of the vectors used to represent user profiles and ence. ant nguistic values, then we define the nd each p is the estimated degree for the user about his her cosine measure in a 2-tuple linguistic context. Given two vectors of nce of the resource x, with respect to x 2-tuple linguistic values
the users provide their preferences about these resources by means of an incomplete fuzzy preference relation. Furthermore, according to results presented in [2], it is enough that the users provide only a row of the preference relation. Then, we use the method proposed in [2] to complete the relations. Once the system completes the fuzzy linguistic preference relation provided by the user, it is possible to obtain a vector representing the user preferences on the topics of interest. Next, we explain this process in detail: (1) Acquiring the user preferences on a limited number of research resources: At the beginning, the main goal is to help the users to provide their preferences assuring that these preferences are as consistent as possible. The system shows users the five most representative resources, R ¼ fr1 ... ; r5g, and asks them to express their preferences by means of an incomplete fuzzy linguistic preference relation (see Fig. 4). The users only fill those preferences that they wish, assigning labels of S4. In the preference relation, each preference value pij represents the linguistic preference degree of resource i over the resource j according to the user feeling. As aforementioned, the simplest case would be to provide a relation with only one row of preference values: P ¼ p12 p13 p14 p15 x xxx x x x x xx x x xx x x 0 BBBBBB@ 1 CCCCCCA : ð6Þ Then, the system completes the preference relation P using the method proposed in [2], and obtains the relation P : P ¼ p12 p13 p14 p15 p 21 p 23 p 24 p 25 p 31 p 32 p 34 p 35 p 41 p 42 p 43 p 45 p 51 p 52 p 53 p 54 0 BBBBBB@ 1 CCCCCCA ; ð7Þ where p1j 2 S4 are the degrees inserted by the user about the preferences of the resource x1 with respect to xj, pii represents indifference, and each p ij is the estimated degree for the user about his/her preference of the resource xi with respect to xj. (2) In order to obtain user preferences on topic of interest, i.e., user preference vector, firstly we calculate the user preference degrees on each considered resource according to the preference relation P , and secondly, we use this preference degrees together with the vectors that represent each research resource to obtain the user preference vector. The preference degrees coincides with the dominance degrees of a linguistic preference relation [19]. To obtain them we propose the application of the arithmetic mean xe (Definition 1). Then, the preference degree of the resource i for the expert called DGi, is computed as follows: DGi ¼ xe p i1; ... ; p i5 : ð8Þ Then, to obtain the user preference vector x, i.e. VUx ¼ ðVUx1; VUx2; ... ; VUx25Þ, from the aggregation of the vectors that represents the characteristics of the chosen research resources, i.e., fVR1; ... ; VR5g, weighted by means of the user preference degrees fDG1; ... ; DG5g. To do that, we use the linguistic weighted average operator defined in Definition 3, and then each position k ¼ f1; ... ; 25g of the vector VUx, is computed as follows: VUxk ¼ xw l ½ ð Þ VR1k;DG1 ; ... ; ð Þ VR5k;DG5 : ð9Þ On the other hand, to complete the user profile, the system asks every user to express his/her collaboration preferences, i.e., if he/ she wants to receive recommendations on collaboration possibilities with others users. This could help users to develop multi-disciplinar works or participate in collaborative research projects [46]. They should respond to this question with ‘‘Yes” or ‘‘No”. 3.2. Recommendation strategy In this phase the system generates the recommendations to deliver the information resources to the fitting users. This process is based on a matching process developed between user profiles and resource representations [17,36]. To do that, we can use different kinds of similarity measures, such as euclidean distance or Cosine Measure. Particularly, we use the standard cosine measure [36]. As the components of the vectors used to represent user profiles and research resources are 2-tuple linguistic values, then we define the cosine measure in a 2-tuple linguistic context. Given two vectors of 2-tuple linguistic values: Fig. 3. Interface to define the disciplines of the resource scope. 36 C. Porcel, E. Herrera-Viedma / Knowledge-Based Systems 23 (2010) 32–39