An adaptive videos enrichment system based on decision trees for people with sensory disabilities José Francisco Saray Villamizar Université de Lyon, CNRS Université Lyon 1, LIRIS, UMR5205, F-69622, France jsaray@gmail.com Benoît Encelle Université de Lyon, CNRS Université Lyon 1, LIRIS, UMR5205, F-69622, France bencelle@liris.cnrs.fr Yannick Prié Université de Lyon, CNRS Université Lyon 1, LIRIS, UMR5205, F-69622, France yprie@liris.cnrs.fr Pierre-Antoine Champin Université de Lyon, CNRS Université Lyon 1, LIRIS, UMR5205, F-69622, France pchampin@liris.cnrs.fr ABSTRACT The ACAV project aims to improve videos accessibility on the Web for people with sensory disabilities. For this purpose, additional descriptions of key visual/audio information of the video, that cannot be perceived, are presented using accessible output modalities. However, personalization mechanisms are necessary to adapt these descriptions and their presentations according to user interests and cognitive/physical capabilities. In this article, we introduce the concepts needed for personalization and an adaptive personalization mechanism of descriptions and associated presentations is proposed and evaluated. K e ywords adaptive videos enrichment system, videos accessibility, videos annotation, decision trees, people with sensory disabilities. 1. INTRODUCTION The amount of Web videos is continually growing up [17] and, as a result, challenges a lot of accessibility problems for people with disabilities [3][17]. The ACAV project [15] aims to explore how accessibility of Web videos can be improved and tackles two research questions [15] i) what i s r equir ed t o make a video accessibl e on t he Web and how c an i t be achi eved?, and ii) how t o inc r ease t he number of accessibl e videos on t he Web? The ACAV approach consists in providing a tool for describing key visual/audio elements of a given video and another tool for presenting these descriptions during the playing of the video in an accessible way (according to the user disabilities) [15]. The work outlined in this article contributes to the development of the second tool: the question we address is how to provide r e l e vant desc riptions and r el e vant desc riptions pr ese ntations to a give n use r during the visualization of a video? Section 2 formally describes the research problem we tackle. Section 3 introduces the adaptive personalization mechanism we use and Section 4 evaluates this mechanism. Section 5 deals with related works. Finally, we discuss our work and highlight some future work in Section 6. 2. PROBLEM DESCRIPTION Improving videos accessibility for people with visual and hearing impairments requires video annot at ions Î i.e. additional descriptions (electronic texts) about key visual or audio elements attached to temporal intervals of the video [13]. For instance, according to [9], several types of visual elements have to be described: information about settings, actions, etc. Îwith as many complexities as possible [6] (hereafter called Level s of Det ail (LoD) - g0i0 hqt c ugvvkpi cppqvcvkqp. ÐjqwugÑ yknn jcxg c NqF qh 3. Ðjqwug ykvj vyq tqqouÑ will have a LoD of 2, etc.). As a result, an annotation contains a t yped textual description of a given LoD and can be presented using a given out put modali t y [15]: e.g. using a female synthetic voice (Text-To-Speech), or using a refreshable Braille display (with regular or contracted Braille). A video with the presentations of its associated annotations is called an enri ched video. However, predefined presentations of annotations for a given enriched video may not fully satisfy the needs of each kind of users: e.g. some visual impaired may want deeper details about characters and settings while others want to get details about actions. Some read Braille and others not. An adaptation mechanism, capable of transforming the presentations of annotations during video visualization is thus needed. Generally speaking, two distinguishable approaches for performing adaptation exist: adaptive and adaptable ones [4]. ÐRwtgÑ cdaptable approach is not really a good option for us because the end-user has to assume the adaption mechanism by setting up explicitly her preferences before playing the video. On the other hand, pure adaptive approach offers automated adaptation based on the analysis of the user behavior and user- system interaction traces, what better suits our needs. Our approach is close to an adaptive mechanism: the user interacts with the system through common actions in a video player (e.g. the PLAY/PAUSE events) and through new actions. As an example, a ÐFEEDBACKÑ event - fired by a spacebar key press, is used to indicate to the system when the user dislikes the presentation(s) of annotation(s) (i.e. r ej ect ed if event is fired otherwise accept ed by the user). 2. 1 Formal d esc ription: d e finitions Pr ese ntation attribut es: Presentation attributes PA ={A 1 .È. A n } discussed in [19] are attributes concerning an annotation presentation. For our case, considered presentation attributes are annotation t ype (e.g settings, actions), out put modali t y (e.g. Braille or Text To Speech (TTS)) and its LoD. Domain of a pr ese ntation attribut e : for a presentation attribute A i we define the attribute domain Dom(A i ) as the set of discrete values {a 1 , a 2 , ..., a n } we can choose Ai from. For example if A i = "Modality ", a possible Dom(A i ) will be {'Braille', 'TTS', etc.}. Annotation pr ese ntation: an annotation presentation P is a set of values {V 1 ,V 2 .È.X n } for presentation attributes {A 1 ,A 2 .È.C n } such that V i Dom(A i ) . Permission to make digital or hard copies of all or part of this work for personal or classroom use is granted without fee provided that copies are not made or distributed for profit or commercial advantage and that copies bear this notice and the full citation on the first page. To copy otherwise, or republish, to post on servers or to redistribute to lists, requires prior specific permission and/or a fee. W4A2011 Î Communi c a t ions, March 28-29, 2011, Hyderabad, India. Co-Located with the 20th International World Wide Web Conference. Copyright 2011 ACM 978-1-4503-0476-4...$5.00.