Haut PDF Leveraging social relevance : using social networks to enhance literature access and microblog search

Leveraging social relevance : using social networks to enhance literature access and microblog search

Leveraging social relevance : using social networks to enhance literature access and microblog search

community. Besides, InRank and DiscussRank show also intersting results with some notable microbloggers in respect to query topic are ranked in the top results. 3. Relevance definition in social networking environments may integrate differ- ent factors. In microblogging context, we propose to integrate the topical rel- evance, the social relevance, and the temporal relevance. Topical relevance is defined by the textual similarity between the tweet and the query. The social relevance is defined by the position of the microblogger in the social network particularly his influence and leadership as proposed by InfRank, LeadRank and DiscussRank algorithms. In particular, we propose to integrate these relevance factors within a Bayesian network model. Two Bayesian network topologies for microblog search model are introduced in the work including inference Bayesian network and Belief Bayesian networks. Conducted exper- iments on TREC Microblog Track 2011 and 2012 show that social networks and temporal features may enhance the real-time search within microblogs in particular for socially and temporally queries. Moreover, we conclude that term frequencies are not more informative in comparative to simple pres- ence of query terms in the tweet which is a good of indicator of relevance. Tuning experiments for temporal interval ∆t show that best retrieval preci- sion are achieved with ∆t value is near to 1day. Comparing the impact of social importance algorithms, we note slight improvement realized by lead- rank algorithm compared to InRank and DiscussRank. Leadership may thus express the social importance of tweets in the context of real-time search. Meanwhile, the performances of these algorithms are limited by evaluation constraints where retweets are presumed irrelevant. Furthermore, we learn from experiments that analyzing the temporal distribution of terms configu- rations instead of single terms of all the query terms is more useful to detect activity period of the query topic. In fact, this method overcomes the prob- lem of popular terms with consistently high distributions over time. Final results show that our belief Bayesian network model present better results than the Bayesian network model. Although the precisions of our models are quite low compared to first ranked systems in TREC Microblog Track 2011, our belief Bayesian network model overpasses the first system in Microblog Track 2012 with an improvement of 23%.
En savoir plus

216 En savoir plus

An exploratory study on using social information networks for flexible literature access

An exploratory study on using social information networks for flexible literature access

after the analysis of the most downloaded documents’ content. Indeed, we high- light that most downloaded documents are published recently (2008) and re- lated queries contain mainly specific terms belonging to new research topics (social IR, collaborative IR etc). Unlike most cited document’s related queries, results set contain restricted number of documents dealing with few topics and authored by a restricted group of authors that usually work together. Authors of most-downloaded documents are central entities in their neighborhood or re- search topic and already have several collaborations in the same research topic, consequently they have a higher closeness value. In addition, authors of most downloaded documents are not authoritative in the whole data collection but in their neighborhood and they have the opportunity to be cited thanks to their past published documents. This explains the good results given by the pagerank and hits measures too.
En savoir plus

12 En savoir plus

Fresh and Diverse Social Signals: Any Impacts on Search?

Fresh and Diverse Social Signals: Any Impacts on Search?

In this paper, we exploit novel social characteristics based on the same principle that some related work. However, un- like previous work [3, 7, 11, 12, 19, 23, 25], that attempt to improve a search on specific social networks (e.g. YouTube, Twitter) by exploiting their own signals, our work specifi- cally focuses on, firstly, exploiting various signals from differ- ent sources as document priors to enhance Web IR. Secondly, considering diversity of signals as an additional factors in the estimation of the resource relevance. We note that in related work diversity has been applied only to the textual content of the document [2, 32]. Thirdly, evaluating the impact of the freshness of signals on the search performance by using their creation date. Fourthly, normalizing the distribution of signals on the resource using the age of the resource. Fur-
En savoir plus

11 En savoir plus

Diversity and Popularity in Social Networks

Diversity and Popularity in Social Networks

Two other recent papers have started to tackle these issues. 1 Currarini, Jackson, and Pin (2008) study a matching process of friendship formation. They document several empirical patterns of homophily and explain them through a combination of biases with respect to choice (preferences) and chance (opportunities presented by the matching process). By design, their model does not allow degree to vary across individuals. This makes their entire analysis quite different from ours. Interestingly, however, a number of their predictions are also supported by our model. Jackson (2008) incorporates homophily into the random graph model of Chung & Lu (2002a,b). 2 Again by design, homophily cannot vary with degree in this approach. Also, degree distributions constitute an outcome of our model while they are an input of Jackson (2008). Thus, our analysis and these two papers study homophily patterns in networks from three distinct and rather complementary points of view. In particular, we provide the first study of the relationship between homophily and individual’s degree.
En savoir plus

36 En savoir plus

Trust and Manipulation in Social Networks

Trust and Manipulation in Social Networks

c CEREC, Université Saint-Louis –Bruxelles, Belgium. September 18, 2013 Abstract We investigate the role of manipulation in a model of opinion formation where agents have opinions about some common question of interest. Agents repeatedly communicate with their neighbors in the social network, can exert some e¤ort to manipulate the trust of others, and update their opinions taking weighted averages of neighbors’opinions. The incentives to manipulate are given by the agents’pref- erences. We show that manipulation can modify the trust structure and lead to a connected society, and thus, make the society reaching a consensus. Manipulation fosters opinion leadership, but the manipulated agent may even gain in‡uence on the long-run opinions. In su¢ ciently homophilic societies, manipulation accelerates (slows down) convergence if it decreases (increases) homophily. Finally, we investi- gate the tension between information aggregation and spread of misinformation. We …nd that if the ability of the manipulating agent is weak and the agents underselling (overselling) their information gain (lose) overall in‡uence, then manipulation re- duces misinformation and agents converge jointly to more accurate opinions about some underlying true state.
En savoir plus

39 En savoir plus

Influence and Social Tragedy in Networks

Influence and Social Tragedy in Networks

The main motivation of the model is collective wasteful behavior towards goods such as power, prestige or status. The welfare losses caused by con- sumption of such goods have been established by Hirsch (1976) and Frank (1985) and is clearly demonstrated by increased investments made, in some occasions, by rival agents: from the early 1950s through the mid-1960s, the US brewing industry was involved in a “ game of market power ”, the advertis- ing spending per barrel rising from $5.00 in 1950 to $8.10 in 1963 (Tremblay and Tremblay, 2007, p. 68); for several years now, US colleges and universities are clearly engaged in a “ game of educational prestige ”, seeking to attract the best students through increased spending or reduced price (Winston, 2000). A structural analysis of collective wasteful behavior towards status goes back at least to Thorstein Veblen, who saw the consumption of some goods or services as “ conspicuous expenditures ” driven by “ the stimulus of an invidious comparison which prompts us to outdo those with whom we are in the habit of classing ourselves ” (Veblen, 1899, p. 103). Interestingly, Veblen emphasized the role of the hierarchic structure of social classes (di- rected network) on the diffusion of the pecuniary standard of living (wasteful behavior).
En savoir plus

25 En savoir plus

Computing in Social Networks

Computing in Social Networks

Computing in Social Networks 3 1 Introduction The past few years have witnessed an explosion of online social networks and the number of users of such networks is still growing regularly by the day, e.g. Facebook boasts by now more than 400 millions users. These networks constitute huge live platforms that are exploited in many ways, from conducting polls about political tendencies to gathering thousands of students around an evening drink. It is clearly appealing to perform large-scale general purpose computations on such platforms and one might be tempted to use a central authority for that, namely one provided by the company orchestrating the social network. Yet, this poses several privacy problems, besides scalability. For instance, there is no guarantee that Facebook will not make any commercial usage of the personal information of its users. In 2009, Facebook tried to change its privacy policy to impose new terms of use, granting the company a perpetual ownership of personal contents – even if the users decide to delete their account. The new policy was not adopted eventually, but highlighted the eagerness of such companies to use personal and sensitive information.
En savoir plus

18 En savoir plus

Computing in Social Networks

Computing in Social Networks

Secure computing in a Social network. Whereas scalability characterizes the spatial, computational and message complexity of the computation, the secure aspect of S 3 encompasses accuracy and privacy. Accuracy refers to the robustness of the computation and aims at ensuring accurate results in the presence of dishonest participants. This is crucial in a distributed scheme where dishonest participants might, besides disrupting their own input, also disrupt any intermediary result for which they are responsible. The main challenge is to limit the amount of bias caused by dishonest participants. Privacy is characterized by the amount of information on the inputs disclosed to other nodes by the computation. Intuitively, achieving all three requirements seems impossible. Clearly, tolerating dishonest players and ensuring privacy calls for cryptographic primitives. Yet, cryptographic schemes, typically used for multi- party computations, involve too high a computation overhead and rely on higher mathematics and the intractability of certain computations [2, 3, 4]. Instead, we leverage users’ concern for reputation using an information theoretical approach and alleviate the need for cryptographic primitives. A characteristic of the social network context is indeed that the nodes are in fact users who might not want to reveal their input, nor expose their misbehavior if any. This reputation concern, as illustrated in the figure below, determines the extent to which dishonest nodes act: up to the point where their misbehavior remains discrete enough not to be discovered. In a system where users report on the misbehaviors they detect, dishonest node might be tempted to issue spurious reports on other users. However, in the context of social networks, two key factors help thwarting such a threat: First, reports are intended to be read by users (not programs) who can assess the credibility of the reports and decide whether to take them into account; Second, the knowledge of the social ties between users can be leveraged. For instance, reports from an enemy or a joint report issued by users that are connected in the social network could be disregarded. Such techniques proved e fficient in areas as diverse as on-line games [5], recommendation systems [6], and spam filtering [7].
En savoir plus

17 En savoir plus

Access to European Grey Literature

Access to European Grey Literature

The first case is a kind of legal deposit of grey items. One of the three special scientific libraries in Germany, the German National Library of Science and Technology in Hanover (TIB) celebrated its 50th anniversary in June 2009. TIB defines itself as a transfer centre for scientific knowledge; its task is ―to comprehensively acquire and archive literature from around the world pertaining to engineering and the natural sciences‖. The library places a particular emphasis on the acquisition of grey literature (conference proceedings, research reports, standards and dissertations in print and digital format). Its grey holdings are unique in Germany. In 2010, TIB holds more than 210.000 print and 30.000 digital German research reports on engineering or natural sciences. Each month, around 200 new electronic and 500 print reports are added. TIB is the deposit library of the (digital) final project reports funded by the Federal Ministry of Research. Since 1996, any
En savoir plus

11 En savoir plus

Mining Social Networks and their visual Semantics from Social Photos.

Mining Social Networks and their visual Semantics from Social Photos.

Mining social networks and their visual semantics From social photos 11 More precisely, the simple force considers the number of times a couple appears on different photos. This social force takes into account the number of object-concepts (groups of photos with the same people). Obviously the bride and the groom are the heroes: their nodes are very close and most of the liaisons start from them. Members of the close family are most of the time captured with them (links only with the bride and the groom) and sometimes together either with or without the bride and the groom. The other people are passive participants that are less often on photos and consequently show no liaisons (at least on the reduced graph which is not the case on the original graph). With the proximity force the more a couple is isolated on a photo, the stronger it is. A couple lost among many other people is less prone to social relations. It is surprising that the resulting network looks very much the same as the simple force network. This is a result which was not expected. It suggests the idea that a strong couple is bound to be present on a photo on its own or with few people. It can be noticed that both forces take into account the number of concepts as their denominator.
En savoir plus

21 En savoir plus

Dynamic competition over social networks Dynamic competition over social networks

Dynamic competition over social networks Dynamic competition over social networks

Hence, a strategic influencer faces a number of trade-off. At each stage, he must first choose whether to influence an easily influenceable agent, with low D(k), or a central agent, which might however have a high D(k) and be harder to influence (in particular in the case of remark 3.1 above). Second, he must choose whether to confront the other influencer by choosing the same target or to shield away by choosing another target or playing a mixed strategy (e.g. if he has a much weaker influence potential than his opponent). Both trade-offs are already present in the static targeting problem where the influenced agent is fixed through time (Bimpikis et al., 2016; Grabisch et al., 2017). The key issue when strategic influencers can use dynamic strategies is the trade-off between adopting a forward-looking or a backward-looking perspective. Influencers adopt a purely forward looking perspective if they only focus on the forward diffusion of their influence via the social network and neglect their opponent’s previous actions. They adopt a purely backward looking perspective if their sole focus is to prevent the spread through the network of the influence precedently exerted by their opponent. In general, they ought to use a strategy, which implements a trade-off between both perspectives. We shall analyze these issues by formalizing the problem as a two-player stochastic game.
En savoir plus

25 En savoir plus

Trend detection in social networks using Hawkes processes

Trend detection in social networks using Hawkes processes

The proposed method bares some resemblance with classi- cal works on trend detection. For example, as mentioned above, our algorithm uses Hawkes processes [15], [16] to model the broadcasting/posting times of messages in a social network, which is similar to the infinite-state automaton approach of Kleinberg [1]; the difference between both approaches is how to deal with the intensity stemming from the broadcasting activity: while Kleinberg searches the periods in time with a high frequency of broadcasts about similar contents, we study a Hawkes intensity for broadcasts about contents that can increase even by broadcasted messages about different ones. Since the influences of users and topics in our information diffusion Hawkes model generates correlation in broadcasts between different contents, our work also relates through the underlying Hawkes intensity to the work of Wang et al. [2], where the authors propose a probabilistic algorithm that discovers correlated bursty patterns and their periods across text streams; the main difference besides the underlying information diffusion model is that we assume the broadcasts to be about specific predefined topics, whereas Wang et al. use text mining techniques to unravel the topics, defined as probabilities over vocabularies.
En savoir plus

9 En savoir plus

Computing in Social Networks

Computing in Social Networks

Solving the S 3 problem is challenging, despite leveraging this reputation con- cern: to ensure privacy, an algorithm must ensure that even when all the nodes except one have the same inputs, the information obtained by the coalition of faulty nodes cannot know which non-faulty node had a different input. This requires the existence of two configurations of inputs that differ for two nodes, which with high probability lead to the same sequence of messages received by the faulty nodes. In turn, this boils down to swapping two nodes’ inputs trans- parently (from the standpoint of the faulty nodes), which is challenging when the protocol needs to be also scalable and accurate. The scalability requirement (i.e., each node communicates with a limited number of nodes) makes it difficult to find a chain of messages that can be swapped transparently between two nodes in the system. The trade-off between privacy and accuracy can be illustrated by the following paradox: on the one hand verifying that nodes do not corrupt the messages they receive (without digital signature) requires the verifier to gather some information about what the verified node received; on the other hand the more the nodes know about the messages exchanged the more the privacy of the nodes is compromised.
En savoir plus

16 En savoir plus

A Priori Relevance Based On Quality and Diversity Of Social Signals

A Priori Relevance Based On Quality and Diversity Of Social Signals

5. CONCLUSION We proposed in this paper a search model based on users’ actions associated with a document. We proposed to es- timate a social priors of a document by considering sig- nals diversity. The proposed model is based on language model that incorporates this a priori knowledge. Experi- mental evaluation conducted on IMDb dataset shows that taking into account these social features in a textual model improves the quality of returned search results. Finally, to investigate the influence of social networks on the quality of signals, we developed a statistical study on the distribution of social signals for each social network in both relevant and irrelevant documents.
En savoir plus

6 En savoir plus

A Priori Relevance Based On Quality and Diversity Of Social Signals

A Priori Relevance Based On Quality and Diversity Of Social Signals

Using language model to estimate the relevance of document D to a query Q . 𝑷 𝑫 is a document prior. 𝑤 𝑖 represents words of query Q . ► Signals are grouped according to their property 𝑥 ∈ 𝑃: 𝑃𝑜𝑝𝑢𝑙𝑎𝑟𝑖𝑡𝑦, 𝑅: 𝑅𝑒𝑝𝑢𝑡𝑎𝑡𝑖𝑜𝑛 ► The priors are estimated by a counting of actions 𝑎 𝑖 associated with D .

2 En savoir plus

Autonomous search in a social and ubiquitous Web

Autonomous search in a social and ubiquitous Web

10 The color code values used in our demonstrator correspond to nuances of green and red used by Philips Hue. Browser artifacts serve as facades that allow agents to interact with artifacts discovered at run time in their hypermedia environment as they would interact with any artifact in a local workspace (see Section 2.1 ) on the JaCaMo platform. Browser artifacts are instantiated with IRIs of W3C WoT TDs, and we refer to them as “browser” artifacts because they perform functions similar to Web browsers: they retrieve and parse W3C WoT TDs, expose interaction affordances to agents, and translate agents’ actions to interactions with the Web Thing [ 25 ] being described (e.g., via HTTP or CoAP). Unlike regular JaCaMo artifacts, browser artifacts expose metadata (e.g., about the supported types of actions provided to agents) via observable properties. To perform actions, such as the action of changing the color of a light bulb in Listing 1 (line 14), agents use a generic act operation provided by the browser artifact. The act operation takes as arguments the IRI of the action type to be executed as well as IRIs of any required parameter types specified in the W3C WoT TD used by the browser artifact. If the W3C WoT TD provides multiple hypermedia controls for the same action type, the first hypermedia control is used.
En savoir plus

15 En savoir plus

Emotional Social Signals for Search Ranking

Emotional Social Signals for Search Ranking

2.1 Social Signals Social signals represent one of the most popular UGC (User Gener- ated Content) on the Web. Indeed, the Web pages include buttons of different social networks where users can express whether they support, recommend or dislike content (text, image, video, etc). These buttons which describe social activities’ actions (e.g., like, share, +1, etc) are related to specific social networks (e.g., Facebook, Google+, etc) with counters indicating the rate of interaction with the Web resource. In February 2016, Facebook introduces additional emotional signals (reactions), allowing users to interact with posts (resources) across love, haha, wow, angry, and sad (see Figure 1). These reactions are an extension of the like button, to give users more ways to express their feelings towards a post in a quick and easy way. The goal of these new signals is to encourage users to react even if the contents are difficult to like as in the case of disas- ters, gloomy news, death, emotion on movie. Table 1 summarizes the most popular signals on social networks.
En savoir plus

5 En savoir plus

Different Approaches to Influence Based on Social Networks and Simple Games

Different Approaches to Influence Based on Social Networks and Simple Games

case, we have defined and studied, in particular, the influence index of a coali- tion on a player, several influence functions, the set of followers and perfect followers, and the kernel of an influence function. The main difference be- tween the two generalized models of influence lies naturally in the definitions of the influence indices. While in the previous model (i.e., the model with a totally ordered set of actions), the influence index has been defined by the sums of some expressions over the particular sets, in the continuum case the sums are replaced by integrals. These integrals are calculated over particular sets of inclination vectors which are of a smaller dimension than the set of n-inclination vectors. We show the equivalence between the influence index of a coalition on a player and the corresponding influence index in which the coalition in question is treated as one player. For a more detailed analysis of this model we refer to Grabisch and Rusinowska (2009c, [17]).
En savoir plus

27 En savoir plus

Analysis of Social Dynamics on FDA Panels Using Social Networks Extracted from Meeting Transcripts

Analysis of Social Dynamics on FDA Panels Using Social Networks Extracted from Meeting Transcripts

Our experience also suggests a relation between voting behavior and linkage patterns. If people who vote the same way also share linguistic attributes, then this suggests that their attention may be directed towards something that drives their decision outcome. This further suggests the possibility of agreement on a relatively small number of reasons for either approval or non-approval. On the other hand, the absence of links between members who vote the same way suggests that there may be a high diversity of reasons for why individuals vote a certain way. In a similar manner to how we define specialty cohesion, we define vote cohesion as the proportion of edges in a graph that connect two panel members who vote the same way. Vote cohesion percentile is the proportion of random graphs, out of 1000 samples, that have lower vote cohesion than a graph representing a given meeting. There are 11 meetings in which there is a voting minority that has at least two people in it. These are used to generate a second meeting-specific distribution found (textured) in Figure 6. This is contrasted against the vote cohesion percentile distribution for 1000 random graphs – a uniform distribution.
En savoir plus

7 En savoir plus

A Priori Relevance Based On Quality and Diversity Of Social Signals

A Priori Relevance Based On Quality and Diversity Of Social Signals

Finally, according to this preliminary statistical study, we observe that each social network has its own specific in- fluence on the quality of its social signals. The quality of signals, provided by Facebook, Twitter, Google+ and Deli- cious, in a document depend on their frequencies, the more the signals are frequent on the resource, the more its a pri- ori importance increases. However, the LinkedIn signal does not only depend on its frequency in the document because it has in itself a power of mature trust compared to the other signals. This amounts to the maturity of LinkedIn users who are well reputed compared to other social networks users.
En savoir plus

5 En savoir plus

Show all 10000 documents...