A Trust-Enhanced Recommender System Application: Moleskiing

A Trust-Enhanced Recommender System Application: Moleskiing
of 6
All materials on our website are shared by users. If you have any questions about copyright issues, please report us to resolve them. We are always happy to assist you.
Related Documents
  See discussions, stats, and author profiles for this publication at: A trust-enhanced recommender systemapplication: Moleskiing Conference Paper  · January 2005 DOI: 10.1145/1066677.1067036 · Source: DBLP CITATIONS 117 READS 63 3 authors: Paolo AvesaniFondazione Bruno Kessler 128   PUBLICATIONS   2,552   CITATIONS   SEE PROFILE Paolo MassaGalliera Hospital 91   PUBLICATIONS   2,562   CITATIONS   SEE PROFILE Roberto TiellaFondazione Bruno Kessler 24   PUBLICATIONS   245   CITATIONS   SEE PROFILE All content following this page was uploaded by Paolo Massa on 24 April 2015. The user has requested enhancement of the downloaded file. All in-text references underlined in blue are added to the srcinal documentand are linked to publications on ResearchGate, letting you access and read them immediately.  A Trust-enhanced Recommender System application:Moleskiing Paolo Avesani ITC/irsTTrento, Italy avesani@itc.itPaolo Massa ITC/irsTTrento, Italy massa@itc.itRoberto Tiella ITC/irsTTrento, Italy ABSTRACT Recommender Systems (RS) suggests to users items theywill like based on their past opinions. Collaborative Filter-ing (CF) is the most used technique to assess user similaritybetween users but very often the sparseness of user profilesprevents the computation. Moreover CF doesn’t take intoaccount the reliability of the other users. In this paper wepresent a real world application, namely, inwhich both of these conditions are critic to deliver person-alized recommendations. A blog oriented architecture col-lects user experiences on ski mountaineering and their opin-ions on other users. Exploitation of Trust Metrics allows topresent only relevant and reliable information according tothe user’s personal point of view of other authors trustwor-thiness. Differently from the notion of authority, we claimthat trustworthiness is a user centered notion that requiresthe computation of personalized metrics. We also presentan open information exchange architecture that makes useof Semantic Web formats to guarantee interoperability be-tween ski mountaineering communities. 1. INTRODUCTION Recommender Systems (RS) suggest to users items they maylike, based on users’ previous opinions. In general, RSs as-sumes all the information is reliable but, in open environ-ments, this assumption is no more true and evaluating thequality of the content provided by users becomes an impor-tant issue.An emergent technique used to deal with the quality as-sessment in open environments is to ask users to explicitlyspecify which other users they trust. For example, on, a site where users can review products, users canalso specify which other users they trust (i.e. “reviewerswhose reviews and ratings they have consistently found tobe valuable” 1 ) and which ones they don’t. Then the user’s 1 From the Web of Trust FAQ( wot)web experience is personalized based on this “web of trust”.Similar patterns can be found in online communities (forexample, in which millions of users posts newsand comments daily), in peer-to-peer networks (where peerscan enter corrupted items), in e-marketplace sites (such and in general in many open publishing commu-nities [5]. These approaches mimic real life situations inwhich it is common habit to rely on opinions of people wetrust.In those computational settings, trust metrics [3, 6, 9] are emerging as a powerful technique. The idea is to use trustpropagation in order to predict the level of trustworthinessin unknown users. These trust scores can then be used topersonalize the user experience by emphasizing content en-tered by trusted users and hiding content provided by unre-liable ones.In this paper we present a community Website whose goal is to make ski mountaineering safer by ex-ploiting information and communication technologies. Userscan share their opinions about the snow conditions of the dif-ferent ski routes and also express how much they trust theother users. Moleskiing is a Recommender Systems poweredwith the use of trust propagation.The contributions of this paper are as follows: •  We present a running application that is conceived tosolve a real problem involving ski mountaineers. •  We argue how the use of a  local   trust metric can beeffective in improving Recommender Systems perfor-mances and propose a preliminary efficient  local   trustmetric. •  We describe Moleskiing open-publishing informationarchitecture in which all the community informationare published in Semantic Web formats. The infor-mation is not required to stay in Moleskiing serversbut can be published wherever on the Web and usersare anyway able to get personalized recommendationsfrom last point is especially important from a research pointof view. Research in Trust Metrics is still in its infancyand we believe one of the reason is the lack of freely avail-able datasets on which trying different solutions and models.  The goal of Moleskiing is hence also to provide an open com-putational framework for studying and evaluating differentTrust Metrics, Recommender Systems and in general Adap-tive Personalization techniques on data from a communityof real users. 2. SKIMOUNTAINEERINGDOMAIN In ski mountaineering, both the ascent and descent of a peakare made entirely on skis, using climbing skins and perhapsski crampons for traction on the ascent, and then descendinga continuous ski route back down to the base.However, ski mountaineering may become very risky. Avalanchesrepresent an ubiquitous hazard that may arise by an erro-neous situation assessment.To know in advance snow conditions plays a key role in per-forming a ski tour safely. However it is difficult to havefirst hand evidence about the snow conditions of every sin-gle route, even for security authorities. As a consequence,in order to prevent or to reduce the avalanches hazard, itis a common pratique for ski mountaineers to share theirexperiences. Lately, it is starting to become common for skimountaineers to publish their comments on the Web and tofoster this sense of community sharing vital information. Infact, these days, a pretty common use case is the follow-ing: firstly, the ski mountaineers tries to find on the Webinformation about the snow conditions of some routes shewould like to experience the day after, especially lookingat other ski mountaineers’ diaries and reports. Secondly,she performs the routes and then, when she’s got back athome safely, she reports on the present snow conditions of the recently experienced route. In this way she contributesback with some fresh and important knowledge to the globalcommunity knowledge.However, since we are dealing with information that canmake the difference between danger and safety, there is anew, huge challenge: assessing the reliability of the ski moun-taineers’ reports.The system we present in this paper has precisely this goal:filtering ski trip annotations based on the trustworthiness of the user who entered them. 3. TRUST-AWARERECOMMENDATIONS In Recommender Systems, the standard way of filtering in-formation in order to personalize it according to the user’sopinions is Collaborative Filtering (CF) [4, 2]. The basic idea is simple: in order to create recommendations for acertain user, CF firstly finds users whose opinions are simi-lar to her and then recommends to her items that were likedin past by those similar users. Typically, CF is used in RSsthat suggest movies, songs, books.However, the domain of Moleskiing is somehow differentfrom a typical CF scenario. In Moleskiing the emphasisis on security and users are invited to “rate” a route alsobased on the present snow conditions and not only abouthow much they do like such a route. In this sense users whorate routes in similar ways don’t have necessarily the sameopinions about how much a route is interesting or worth-while. Moreover, in ski mountaineering domain, the infor-mation becomes old quickly: the most important factors forsecurity are the weather and snow conditions and of coursethey are not expected to remain constant over months. Thisfact exacerbates one of the key weaknesses of CF, data spar-sity, which often makes impossible the first mandatory stepof finding users similar to the current one [7]. This problemis especially evident just after the deployment of the systemwhen no ratings are available. Moreover, since the goal of Moleskiing is to make ski mountaineering experiences safer,a special relevance assumes the reliability of the commentsentered by the users about the routes.These peculiarities makes Moleskiing a bit different fromtypical CF scenarios and hence we have chosen to enhancethe RS with trust-aware techniques. The intuition is the fol-lowing: while users can still rate routes about their currentsecurity level, they can also “rate” other users by elicitinghow much they find their comments and reports useful andaccurate. This is their “web of trust”. Then the systemshows to the user mainly information provided by users shetrusts. However there is a problem of coverage: every useris expected to being able to provide a direct trust state-ment only about a small number of other users. In orderto make use also of the information provided by “unknown”users (users the current user has not issued a trust state-ment about), the system exploits trust propagation to infera personalized trust score also for unknown users. The al-gorithms that predict trust in unknown users based on theglobal trust network are called “trust metrics” and are cov-ered in Section 5.In this way, the cornerstones of our system become thetrust statements issued by users about other users. In or-der to obtain this information more easily, we have adoptedan open publishing architecture, in which the information(trust statements and comments on routes) can be decen-tralized published in Semantic Web [1] formats. In this way,the system can potentially aggregate the information avail-able over every community or even single ski mountaineersand does not restrict itself to the (possibly few) users of A description of the open decentralized ar-chitecture is given in Section 6. 4. THEMOLESKIINGAPPLICATION In Moleskiing, there is a clear distinction between ski routesand ski trips. Ski route refers to the itinerary, while ski tripis the experience of a certain ski mountaineer at a certaindate on a certain ski route. While information about skiroutes is entered by domain experts and is expected to bestatic and certified, ski trips are comments created by usersand hence they are dynamic and, more importantly, they canbe unreliable or even totally inaccurate. Our architecturereflects this dichotomy, in fact the ski routes part is storedon a legacy system while the dynamic one relative to ski tripsmakes use of a blog platform. In particular, every user canmaintain a Moleskiing-hosted blog where she can express,using a simple interface, comments and ratings about herski trips. In a similar way, they can keep a list of users theyknow and express how much they trust them, that is, howmuch they find their ski trips annotations useful and reliable.We will see in Section 6 how it is not required for Moleskiingusers to keep this information on Moleskiing servers but theycan publish it wherever on the Web, possibly in their already  Figure 1: Screenshot of homepage used blog.The typical use case on is the following. Anuser looks for ski trips information in order to decide whichroutes she is going to experience the day after. Special at-tention is deserved to the current snow conditions, reportedby other users. Then she performs the ski trips and, whileat home, she comments the routes specifying the currentsnow conditions and hence adding her fresh experience tothe overall history of that route.In this scenario, the use of trust affects the application inthe first step. In fact, since security is one of the goal of Moleskiing, the system tries to show to every user only reli-able comments and to filter out unreliable ones. Trust scoresof users (and hence reliability of their comments) are definedon a per-user basis and hence can be different from the pointof view of different users. In order to understand which usersare trustworthy from the personal point of view of every sin-gle user, the system runs periodically a local trust metricson the overall trust network. In this way, when an user in-teracts with, the system knows how much theinformation provided by every other user should be takeninto account as useful and reliable. A description of thetrust metric is given in Section 5.The first page of (see Figure 4) presents a listof the routes commented in the last 15 days. Ski trips anno-tated before are simply discarded because information aboutthe snow conditions is already too old. Precisely, only routesthat are rated as secure in the last 15 days by the majorityof “trustable users” (users with a trust score not less than0.6 in the [0,1] interval) can be displayed in the list of rec-ommended routes. These routes are then ranked based onthe average rating given by those users, weighted by theirtrust score. The chronological order is taken into accountas well, since users reports of routes experienced recentlyshould reflect more precisely current snow conditions.The system is designed in a way that users are always invitedto expand and fine tune their web of trust. In fact, it isalways possible to click on a user nickname and to readher blog (containing her routes annotations and her web of trust). In this way, the current user can get a first-handevidence and express trust in this user explicitly. Moreover,an user can always see the list of users not yet rated orderedby predicted trust or see the predicted trust score assignedby the system to every other user. Since the web experienceon is driven by the user’s web of trust, webelieve it is very important that the user has it under controland is invited to update it in order to reflect her real viewsof other ski mountaineers’ trustworthiness. 5. TRUSTPREDICTION Trust metrics [3, 6, 9] are an emerging research topic. A trust network (or social network) is built aggregating all thecommunity trust statements into a single directed weightedgraph. Trust statements are weighted and range from totaldistrust to total trust: for example, in the real interval [0 , 1].Trust statements are also subjective: it is normal to have anuser trusted with different scores by different users. Theyare also asymmetric in the sense that, if A trusts B as 0.9,this does not mean that B has to trust A as 0.9 as well.Given a current user, it is possible to predict trust scoresin users she has not expressed a trust statement about, byexploiting controlled trust propagation. The assumption isthat if user A trusts B at a certain level and B trusts C atanother level, something can be inferred about the level of trust of A in C.An important classification of trust metrics is in global andlocal ones [7, 9]. With local trust metrics, the very personal and subjective views of the user is taken into account whenpredicting trust scores in unknown users. For this reason,the trust score of a certain user can be different when pre-dicted from the point of view of different users. Instead,global trust metrics computes a “reputation” score that ap-proximates how much the community as a whole trusts aspecific user. In this way, global trust metrics don’t takeinto account the subjective opinions of each user but av-erage them into standardized global values. PageRank [8],one of the algorithm behing the search engine Google.comis an example of global metric. In general, while local trustmetrics can be more precise and tailored to the single user’sviews, they are also computationally more expensive, sincethey must be computed for each user whereas global onesare just run once for all the community.Another interesting feature of local trust metrics is the factthey can be attack-resistant [6]: users who are consideredmalicious (from a certain user’s point of view) are excludedfrom trust propagation and they don’t influence the per-sonalization of users who don’t trust them explicitly. Thisfeature is especially important in our domain because thereliability of users (and hence their ski routes annotations)is one of the main concerns. A more complete descriptionof trust metrics and related concepts can be found in [7].We have implemented a preliminary trust metric to be usedin Moleskiing application. Since there are no comparativeevaluations of different trust metrics at present time, wehave chosen to start with this one and to improve it as longas real data starts to become available from users. In future,  we also plan to evaluate and compare this one and otherproposed metrics [3, 6, 9]. The goal of this preliminary metric is to be time-efficient so that computing trust scoresin unknown users for every user does take a limited amountof CPU time.The trust metric, named MoleTrust (ver 0.1), works bywalking the social network starting from the input user andby propagating trust along trust edges. Precisely, the Mo-leTrust trust metric works in 2 steps. The first step’s pur-pose is of destroying cycles. An example of cycle is thefollowing:  A  trusts  B  as 0 . 6,  B  trusts  C   a s 0 . 8,  C   trusts  A as 0 . 3. The problem created by cycles is that they requirepassing over a node many times adjusting progressively thetemporary trust score until this value converges. Insteadwe would like to have a trust metric that is able to walkon every user just once and, in doing this, to compute thedefinitive trust value. In this way, the running time is linearwith the number of nodes.Let us assume that we are predicting trust scores of unknownusers from the point of view of user  Me  . MoleTrust firstlyorders users based on shortest-path distance from user  Me  .A parameter of MoleTrust is the Trust Propagation Hori-zon: trust is not propagated at distances greater than thisvalue (the default is 3). The intuition is that the reliabilityof the propagated trust decreases with every new trust hop.Moreover, in this way, the number of nodes the trust metrichas to consider is reduced and this means smaller compu-tational time. At this point, MoleTrust removes any trustedge from a user at a certain distance users with a lower orequal distance: for example, every edge from users at dis-tance 3 to users at distance 1, 2 or 3 are removed from thesocial network. The first step ends here. The social networkis now a directed acyclic graph where trust flows from  Me   toother users and it never flows back, i.e. there are no cycles.The second step is a simple graph walk over the modifiedsocial network, starting from user  Me  , whose trust score ismaximum by definition. MoleTrust computes first the trustscore of all the users at distance 1, then of all the usersat distance 2, etc. The trust score of one user at distance x  only depends on trust scores of users at distance  x  −  1,that are already computed and definitive. For predicting thetrust score in an user, MoleTrust analyzes the incoming trustedges. However only trust edges coming from users with apredicted trust score greater than 0.6 are considered. Theother users are not trustworthy and their trust statementsshould simply be ignored.The predicted trust score of user  B  is the average of all theincoming trust edge values, weighted by the trust score of the user who has issued this trust statement. The intuitionis that opinions of users about  B  are weighted based on their(explicit or predicted) trustworthiness. Because of the trusthorizon and because of the structure of the network, it ispossible that a trust metric is not able to reach every nodeand to predict its trust score.We have implemented the trust metric extending the opensource Java package “Jung” ( this section we have presented the preliminary trust met-ric used in Moleskiing. We are aware that the first step canremove significant trust links and result in inaccurate trustscore predictions. This is done in order to reduce the com-putational time, in fact in this way there is no need to passon one user more than once to compute her final trust value.We plan to verify if this simple but efficient trust metric issuitable enough for the Moleskiing community or if we needto design more complex trust metrics as soon as we will startcollecting significative amount of data from users. 6. BOOSTINGTRUST Recommender Systems in general works well when therehave a sizeable quantity of information available. This meansthey require a large user base and a large number of anno-tations provided by those users. However, as long as newcommunity sites arise, the potential community of users isfragmented in smaller communities, often non-exchanginginformation. In this way, every system can have access onlyto the information provided by the users of their own (oftensmall) community. As a consequence, every systems’ perfor-mances tend to decrease. Moreover, an user of a communitysite cannot beneficiate of the comments of an user of anothersite.To overcome this problem, we have designed an open anddecentralized information exchange architecture. The goal isto support interoperability among medium size communitiesof ski mountaineering. The idea is that every informationin the system (trust statements and ski trips annotations)is published using some Semantic Web [1] formats.Precisely we have defined a new format (MSSA) that en-codes a ski trip, an annotation made by one user about acertain route at a certain date. One of the properties is <trip_rate>  and its value could be 0 if the user does notrecommend the route based on present snow conditions, and1 otherwise.The other Semantic Web format we use is FOAF (Friend-Of-A-Friend) [3]. A person can encode in her FOAF fileinformation about herself and her social relations with otherpersons. In particular, we use FOAF format extended withthe trust extension presented in [3]. With this extension, theFOAF file creator can also express her level of trust in otherusers about a certain context. The minimum possible valueis 0 and represents distrust and the maximum is 1, totaltrust. In a FOAF file, other users’ FOAF files are identifiedwith the  seeAlso  property and, following these links, it ispossible, for example, to walk the overall ski mountaineeringsocial network and aggregate it in order to have trust metricspredict trust scores in unknown users. Table 1 shows theFOAF file of a typical Moleskiing user.Moleskiing exports the information about every user on theWeb: the ski trip annotations in MSSA embedded in the userblog entries and the trust statements in the user FOAF file.Other ski mountaineering communities sites (, for instance) are in the process of doing thesame, creating a sort of open federation. In this way, everysites beneficiate of the user base of the other ones in orderto provide better recommendations and personalization.However, Moleskiing does not require users to create a login
Related Search
We Need Your Support
Thank you for visiting our website and your interest in our free products and services. We are nonprofit website to share and download documents. To the running of this website, we need your help to support us.

Thanks to everyone for your continued support.

No, Thanks