AVIC ( Assistant Virtual Intelligent Cooperative)
AVIC is an application that, integrated with a search engine, allows a reorganization of the results from the query executed on the same search engine, based on the approval (implicit and / or explicit) expressed by the persons mentioned as belonging to the social network of the user (both personal, es. facebook, and vocational education, es. LinkedIn, business organization, etc.).
Ultimately, the application allows you to restructure the ranking associated with the results of a search based on the provided feedback (implicitly or explicitly) from users belonging to the same social network: the results which are deemed valid by trusted users will be privileged and then shown in the first positions of the results.
The objective is to obtain in the top positions what is really interesting for the user and not pages and pages of "noise". Eg., Seeking material on intelligence, you get the first results related to "business intelligence" if your social network is mainly populated by computer, while if the persons to whom you are connected, are dealing with investigations can be obtained before the results related to intelligence as a technique of investigation.
AVIC uses a search engine to obtain preliminary results, (eg. Google, Fast, Exalead) and then rearrange them (re-ranking) on the basis of a specific algorithm that allows not only to bring up or down the pages liked by the social network, but also to identify the social network is more reliable, or less, as a source of suggestions.
This means that the research reaches a high level of customization, as AVIC, as the self-learner system, as well as learn about the interests of the user who’s making the research also takes into account the inputs resulting from feedbacks of the satisfaction expressed by Reliable and close users to the user in question.
Objectives
The AVIC project aims to improve the experience of people who every day have to search for information in the immense amount of documents on Internet. The problem of obtaining reliable results, with respect to the executed searches, now afflicts every web user that uses a search engine: in fact, is often the need to refine the questions query to obtain more meaningful and relevant results.
AVIC intends to focus on the individual, with his interests, his social relations and its habits, so as to give results closer to the "socio-cultural context" of the user, rather than leave them to simple mechanisms for the classification of web pages. Consequently, this study tries to find a partial solution that let the user take an active role through the completion of several actions (rating, bookmarking, and so on). In this way, a search will not be a cool tool to query a web engine, but at the same query correspond clearly different results, determined by the person who conducted the search and in what time.
Techniques
In order to give back to the user the most interesting results respect to their own interests AVIC need for accurate profiling of the individual. Such profiling is implemented either with explicit mechanisms or using implicit profiling techniques.
For an explicit profiling there are several mechanisms:
- Feedback "the gladiator": this expression shows the classic mechanism for expressing the popularity of a webpage (or other resource).
- Tagging of pages: often users enter the sites, to which have more interest in the bookmarks of the browser properly tagged.
- Registration Data: is the classical form of registration adopted in a variety of websites, through which sensitive information is obtained regarding user’s age, the affiliation town, the profession…
Concerning the AVIC implicit profiling, it takes into account several indicators, some of which are "immediately detectable" this means that they are determined instantaneously by the actions of the user. Instead, other inferences about the user interests are the result of reasoning developed in separate processes involving mainly the individuals’ social network.
Among the most immediate detectors is possible to consider the following: - Monitoring and click download: it is assumed that the user finds most interesting a web page with which it interacts by means of mouse clicks (for example, to follow a hyperlink to another page) or activating the download (ex download a scientific paper, music, video, photos, etc..).
- Calculation time spent: it’s useful to note the time spent by the user on the web page, or on a specific domain
- History Navigation: especially in the early stages of user’s profiling, it may be meaningful to look at the history of the browser.
- Bookmarks: as with history, may be useful to get the bookmarks already entered by the user. Remember that the bookmarks are related to the system of tagging explicitly mentioned in the explicit profiling, and the Union of these two elements can lead to a kind of social Delicious.
- Display frequency: discovering the frequency with which a user returns to certain web pages is useful to understand its interests.
.
.
.
Among the implicit profiling techniques there have been also evaluated several approaches that require the reasoning processes on the social network to which the individual belongs. Given that these mechanisms do not have a close dependence with the actions of quotas of the user (for example, the research), the inference procedures are carried out in separate processes (initiated in arbitrary temporal windows).
Some indicators should be taken into considerations concerning the reasoning mechanisms:
- The friends’ nodes:in the current context the user has a network of friends. Under the assumption that the individual may develop similar interests with its closest friends, consequently it leads to keep track of the activities of the user’s closest modes, such as visited web sites, searches made, the tagged pages or valued by feedback)...
- The friends of friend’s modes:thus going further into the network of friendship and consider the friend of friend. For example, you may consider navigating the network of friendships in a given node to the second level of depth.
- Friends in common:between the various exposed indicators, some are more important than others, therefore it is plausible to consider the most interesting nodes, displaying a greater number of friends in common with a particular user.
- Intensity of relations:It begins with understanding the intensity of the relationship that a user has with each node of its social network. Whether the link is a business type, rather than sentimental, it is useful to provide more accurate weights to the resources sought from their neighbors, for example, if the user wants to find a scientific document, it makes more sense to first investigate the network of their colleagues rather than in that of his fun friends.
The various indicators have of course a different weight in the user profile (either implicit or explicit). Therefore, the system assigns a different numerical weight depending on the profiling system taken into account (for example, the documents "suggested" by a mode friend at a distance 1 of the user will have a more influential weight than the resources of mode friend at a distance 2.
From these considerations it is understandable that weighing has a dynamic character and it is necessary to make continuous assessments on the basis of changes in the social context.
.
.
.
Architecture
The system consists of three layers distributed between clients and servers. The client-side component is responsible for collecting the information explicitly given by the user (feedback, tagging, and data recording) and to capture the set of immediate indicators, useful for an implicit profiling (monitoring and click download, the permanence calculation time etc.). Therefore it was made a plug-in for Firefox to be installed in the user’s browser. The plug-in has also a navigation bar that allows making research, releasing feedback and assigning tag about the displayed web page.
The System Server side performs the following tasks:
- collect, store and process the information sent by the plug-in;
- store and update the entire social network;
- implementing the reasoning processes on the individuals’ social network, and on the social network as a whole;
- implementing the processes of data elaboration in response to the user’s search requests;
- stores and updates the list of resources associated with the user;
- implements the processes of assigning weights to the resources associated with the individual(or a group of people from the social network).
Regarding the social network management system the persistent system used is a semantic database. The social network is represented by ontology. The ontology taken into account is FOAF (the Friend of a Friend), which provides a range of concepts to describe the relationships between individuals. Given the expressive limitation of the FOAF ontology, it was necessary extending it with new concepts and relations in order to manage trust relationships between users and between users and documents.