DES804M2 – Design Entrepreneurship Awarenes
NewsViz:
Emotion Extraction from News Articles for Visualisation
E.D. Hanser, Dipl. Des. (FH)
School of Computing and Intelligent Systems
Faculty of Engineering
University of Ulster, Magee Campus
hanser-e@ulster.ac.uk
Abstract
Online news is presented plainly and formally even though its content is mostly very emotional. The NewsViz system proposed here visualises these emotional aspects in thirty second long Flash-animations integrated on the article’s website. The application interprets news texts and automatically creates visualisations. This paper concentrates on the construction of the emotion extraction component of NewsViz which is responsible for abstract background visualisation. The developed Flash prototype detects emotions from news reports and produces initial animations which can be further edited by a Flash-designer. The original text is part-of-speech tagged, adjectives and/or nouns are filtered out and are labelled with an emotion and intensity value. Subsequent identical emotions are combined to longer lasting moods and matched with appropriate animation presets. Different linguistic analysis methods were tested on the prototype. The NewsViz outcome proved to be viable for the limited topic area grasping the overall moods and partly more detailed emotions precisely. NewsViz introduces a novel approach of a universally applicable emotion scheme which is not influenced or biased by an agents character and offers an efficient technique to handle the production of a great number of daily updated news stories. Further development may refine the detection of emotion shifts through summarisation with a dependency parser and the full implementation of football knowledge and common linguistic knowledge. Future work will reveal whether NewsViz is a feasible system when extended to different domains. User evaluations of the NewsViz animations proved enjoyment and appreciation of the new form of news presentation by website visitors. In comparison to similar story visualisation systems, NewsViz uses natural language rather than a regular language for the story description, it offers the option to edit the output and fills the gap of lacking information for background or environment depiction.
Keywords
Natural Language Processing, Semantic Interpretation, Emotion Extraction, Automatic Visualisation
1 Introduction
Text-to-visual mapping relates to the research area of natural language processing which attempts to enable computers to interpret and generate natural human language. Therefore text-to-visual mapping starts with the linguistic analysis of the text. Despite variability, ambiguity and imprecision syntactic tools achieve highly accurate results. Part-of-speech taggers are trainable and reach up to 97% accuracy [UCREL, 2006] when identifying grammatical types of words like noun, verb, adjective and so forth. Dependency parsers determine the subject, object, action relation between words of a sentence. Semantic analysis and actually understanding the meaning of a text is more difficult, because it depends largely on common sense knowledge. Mental images need to be structured, related through logical rules and entered into databases before computational text interpretation is possible. Systems attempting automated story visualisation deliver amazing results for static object and action depiction, but only use restricted language as in the WordsEye system [Coyne and Sproat, 2001]. Automated animation is progressing as well, camera and character animation and interaction is realised in Mario [Friedman, 2003] and CONFUCIUS [Ma, 2006]. Emotion detection and modelling derived from visual, textual or audio stimulus lets computer agents have believable human-like facial expression and gestures. Character personality and long-lasting moods influence momentary emotions achieving coherence.
The above mentioned solutions under perform in extracting information from text for background visualisation which is often not directly stated. Emotion visualisation is only aimed at agent animation and biased through its character and intentions. News stories are loaded with emotion, but presented as formal, neutral facts in online newspapers. Imagery is an important medium for communication directly conveying emotional messages. A requirements survey conducted during this project revealed that many people would like to see more video on the internet.
The proposed software package, NewsViz, automatically creates animations from online news articles based on computer driven text-to-visual mapping. The input for NewsViz is in natural language and is processed by three components: the action visualiser, the emotion and background visualiser and the audio creator. Assuming that action and audio presentation are already available, this paper is devoted to emotion and background visualisation. Here the emphasis of the output lies on expression rather than content reconstruction. Abstract 2D background animations depict emotional aspects filtered from news stories in order to emphasise the mood and atmosphere of articles. The emotion evaluation strictly follows statements in the text and does not add its own interpretation. Adjectives and nouns transport most content and emotion in written texts, thus they are selected and tagged with emotional valence indices. Animations are then built following the list of emotions. The emotion illustration is subtle background animation in order to leave the focus on the action and event visualisation lying in the foreground. Information communication is the main goal. In this prototype of NewsViz action and event depiction is indicated through static, pre-created animation sequences. The aim of NewsViz is to create a more enjoyable form of news presentation which addresses the news reader more directly and personally. NewsViz is a cost-efficient software tool used to build daily updated animations. Testing different language processing methods, including varying word types, sentence or word based evaluation and emotion thresholds revealed that emotion extraction of individual words is mostly successful in NewsViz, but summarisation methods which combine moods create rather unpredictable output and need further improvement. Different news reports seem equally suitable for emotion extraction. Test users reviewed the integration of animations into a news website and considered it a useful gimmick. Further development of NewsViz involves constructing more extensive databases and especially the integration of football and linguistic knowledge.
Related works are discussed in more detail in the next section 2. A thorough description of the prototype can be found in the sections 3 and 4. The testing results are described in section 5. NewsViz is compared to similar systems in section 6. A final evaluation of the NewsViz application and ideas for further development are given in the last section, 7.
2 Background and Related Work
Mutual influence of imagery and text in news presentation has an impact on the readers understanding of articles and requires careful consideration. In addition, multi-modal systems automatically mapping text to visuals face challenges in interpreting human language which is variable, ambiguous, imprecise and relies on common knowledge between the communicators. Enabling a machine to understand a natural language text involves feeding the machine with grammatical structures, semantic relations and visual descriptions to be able to match suitable graphics.
2.1 Current Online News Presentation
News reports are regarded as objective facts represented in a neutral and formal format: a static headline, a summary with one image and the body text eventually with one to three more images, sometimes video highlights are available for download. Indeed storytelling is the most preferred way of communicating as it reaches the audience emotionally, educates and entertains [Gershon and Page, 2001]. Thus reporters found the content of news stories worth mentioning for emotional reasons. But story brevity, scarce background information and the combination of visual and verbal information in news hinders learning by viewers [Graber, 1990]. Emphasis on visual elements is therefore important as they tend to be more memorable than verbal ones.
2. Language Processing Techniques
Computer driven syntactic analysis delivers mostly reliable results. A basic first step is part-of-speech tagging which means identifying grammatical types of words. Software tools like Qtag [Qtag, 2003] attach a tag to each word labelling it as noun, verb, adjective or other. Words with different senses are decided upon statistics derived from trained probabilities. Dependency parsers relate the single words according to their function within a sentence such as subject and object. Semantic interpretation depends on the development of broad databases storing common knowledge to describe mental images. Therefore it is limited to themes for which knowledge exists in digital form. A commonly used tool for determining semantic relations between words is WordNet [Miller, 1995]. WordNet is an extended dictionary collecting word relations such as similarity, hierarchical order, part-of relations or manners. Story Segmentation such as SeLeCT [Stokes, 2003] is an example application based on semantic analysis to find story or subtopic changes within a text. Groups of semantically related words called cohesive ’lexical chains’ are extracted from a text. They are determined through WordNet’s semantic relations and additionally through statistically acquired co-occurrences (e.g. Diego Maradonna, Hand of God). Their starting and end points indicate topical unit boundaries.
2.3 Emotion Systems and Football Knowledge
Sensing emotions from multi-modal input has mainly been researched with the objective to model human-like agents. Becker et al. [Becker et al. 2004] proposed a complex architecture to realistically imitate human emotions for an agent. Long-lasting moods influence momentarily evoked emotions as well as emotions mutually interacting with each other. Emotions are described as combinations of pleasure or arousal and their dominance. Emotions lose valance over time until a lack of stimuli leads to a state of boredom. The football commentary system Byrne [Binsted and Luke, 1999] produces a commentator with emotions. Its emotion generation is built on a compact emotional structure derived from Ekman’s six basic emotions (fear, anger, sadness, happiness, disgust and surprise) [Ekman and Rosenberg, 1997] extended by interest. These sentiments are refined through intensity, target, cause and decay parameters. The agents personality and intentions also interfere with the current emotional state. Football knowledge needed to set preconditions and context and consequently to evaluate events and assign appropriate emotions can automatically be collected from the internet. SOBA [Buitelaar et al. 2006] extracts information from soccer match reports annotes relevant expressions (players, teams, goals, etc.) and generates the knowledge base entities. In contrast, the MoodNews website [MoodNews, 2005] demonstrates a very simple linguistic method to distinguish positive, negative and neutral content in BBC news headlines. It effectively ranks them on a colour scale between good to bad. The three kinds of emotions are appointed through keyword scoring based on a small vocabulary of 160 words and phrases.
2.4 Visualisation Techniques
The Flash video ‘The Unseen Video’ [TheUnseenVideo, 2005] gives a good example of abstract mood visualisation. Weather data of the viewer’s location is automatically retrieved from news websites and influences the look and feel of the animation through shapes, colours and images. In contrast ‘WordsEye’ [Coyne and Sproat, 2001] is devoted to the actual creation of static 3D scenes with objects, actions and environments as described in specially written stories. A database of graphical objects holds 3D models, their attributes, poses, kinematics and spatial relations in low-level specifications. The objects are connected with their linguistic counterparts in an extended WordNet version containing links to their corresponding nouns. CONFUCIUS [Ma, 2006] adds character animation and speech. The ‘Story Picturing Engine’ [Joschi et al. 2004] visualises texts selecting and matching pictures and their annotations from image databases. In ‘Automating Spielberg’ [Friedman, 2003] computer driven visualisation is extended to animation which poses new challenges in setting timing, perspective, camera-angle, zoom, motion sequences and other cinematographic concepts.
The examples in the sections above prove that sufficient subsets of the English language can be mapped to computer understandable language to be used for the visualisation of stories, but still need improvement in precise depiction and animation.
3 NewsViz Design
3.1 Aims and Objectives
NewsViz creates animations from news articles. Abstract design elements show emotions conveyed in the stories. The software package NewsViz and the visuals are presented as reliable and serious. Expression stays within the boundaries of reputable broadcasting and trustworthy reported news articles. Thus emotion extraction is universally applicable and without character bias. The main objective is to provide information. NewsViz is an efficient software tool for designers to be able to build daily updated animations.
3.2 Requirements Analysis
The intended news visualisations produced by NewsViz are aimed at the average news website visitor. This target group involves a wide range of public readership of sophisticated adults aged about 16-60. These users are computer affiliated, access the internet more or less regularly and show an interest in daily events and happenings. Their information offers and sources are imense, so that news needs to be quick and easy to grasp. Conception, content and design has to please the described target audience. The NewsViz software package will be employed by animation designers. They have high visual expectatioins and are familiar with different animation software. NewsViz has to be easy to work with in a busi environment and produce graphically satisfactory results.
A requirements survey for NewsViz distributed among twenty online news readers revealed that 70% of the respondents desire more video animation on news websites. Most readers agree that news stories themselves are emotionally touching, but their content and emotional aspects are only sometimes reflected through imagery. This proves a demand for visualised news stories. Most visitors are willing to spend between fifteen seconds to one minute watching a story preview or even a summary of the whole news article.
Online news readers may desire functionality to stop, play and replay the animations and regulate sound. Short download times are essential for the NewsViz animations on the internet, especially in their roll as previews, because the website visitor otherwise starts reading the article. The animation designer requires functionality to edit the news text, preview the animation, edit the animation and save the project. The text analysis processing and rendering of the animation should be fast and invisible to the designer. Advanced users have options to control these processes. Simple operations should be clearly presented to the user in an easily usable interface.
3.3 System Architecture
NewsViz takes online news articles as input and outputs animations reflecting the content of the news stories. The system consists of three main components: the linguistic analysis, the animation composer and an interface for editing text and animations (Figure 4.1). The linguistic component is separated into three processes which construct different elements for the animation. The emotion extraction tool creates background visuals, the action visualiser depicts people, objects and actions and the audio creator selects music and sound effects. The composer synchronises the different outputs. Here we focus on the development of the emotion extraction tool.
[pic]
Figure 3.1: NewsViz System Architecture
3.4 Emotion Extraction Tool
Original online news articles are inserted into the emotion extraction component (Figure 4.2). Emotional aspects within the news story are identified. These emotions are linked to appropriate presets of background animations.
[pic]
Figure 3.2: Emotion Extraction Component
The first step of processing the natural language text is to determine the grammatical type of all words. The part-of-speech tagger Qtag is utilised for this purpose. This Java based, command line executable program is mainly trained for the English language. It attaches tags to nouns, verbs, adjectives and many more described in its tag set. The tagged text is passed on to the adjective and noun detector. Only these two types of words are selected for further processing since it became obvious during the construction of the emotion index lexicons that adjectives and nouns mainly convey emotional meaning. Similar findings for written texts are described in [Stokes, 2003]. For this reason, this prototype ignores verbs, adverbs and other word types. Following, the emotion word selector checks the words for emotion indices in the extended WordNet dictionary (which is imitated through an XML structure at this stage of the prototype) and affixes emotion tags indicating their kind and intensity of emotion. In relation to football matches three of the seven basic emotions mentioned in [Binsted and Luke, 1999] have been found relevant – happiness, boredom (as a lack of interest or tension) and sadness. Tension is included to specify the tight, nervous and fierce feeling during competitive matches. Words with a neutral emotion index do not describe football relevant emotions. To achieve a coherent course of emotion and animation neutral phases are replaced by the previous mood with decreasing intensity. The intensity value distinguishes three levels – low, medium or high depending on how clearly the word represents an emotion. Emotional words represent short-term emotions. The list of emotion tagged words is handed to the emotion summariser. During the summarisation process subsequent emotions are combined to one longer-lasting mood, if they have the same emotion index. A chronological list of mood chunks is created. Each mood is labelled with its kind, average intensity and the animation duration. The duration is calculated as the percentage of the amount of words belonging to a specific mood to the total number of emotion words. Mood boundaries appear as soon as the emotion of the next word differs. This method is called ‘word by word’ as it compares each detected emotion word with the following one. In order to reduce error and excessive mood swings, the minimum threshold method offers a variable which determines a minimum number of words required to represent an emotion. All moods which are composed of less then X words are removed from the list. Alternatively the sentence based summarisation methods averages assumes that one sentence conveys one idea and consequently one emotion. Hence it calculates an average emotion for each sentence, before combining identical emotions. The mood list is refined through moods discovered with linguistic common knowledge and football facts.
3.5 Databases
The NewsViz prototype accesses four databases: a dictionary, linguistic knowledge, football related knowledge and graphics. The dictionary holds emotion-indices and intensity values of all words. Words which have ambiguous emotional meanings were decided upon their relevance and frequency in football reporting. The database of linguistic knowledge helps to identify emotions in context properly applying common language rules to emotion interpretation. As an example the NOT-rule is called when a negative marker appears in a sentence and turns the next emotion word into its opposite, so the happy tagged word rather expresses sadness. The football knowledge base provides domain related background knowledge. This includes match statistics such as players, teams, referees, nick- or short names, team colours or up to date league tables. It also accommodates game rules or match situation descriptions with their emotional consequences. For example the emotion intensity and combination evoked through a goal depends on the goal difference. If the goal difference is one, this means happiness about the lead, but still enough tension. While a goal leading to a difference of five or greater happiness is very strong, but mixed with boredom. The graphics database contains prefabricated animations which are combinable and adjustable according to mood intensities. They are a collection of .swf-files, which are sorted by mood index and intensity and can be loaded individually into the main movie file.
3.6 Animation Construction
The animation selection component loads the individual animation elements from the graphics database and combines them in a thirty second long movie. Their sequence order, the type of element, the amount of objects loaded and their display time is determined by the weighted mood list. An emotion change in the mood list causes the current animation elements to fade out and to load different elements. Properties like background colour and element size are set according to each mood and its intensity.
[pic]
Figure 3.3: Animations for Happiness, Tension, Boredom and Sadness
3.7 Interface Design and User Interaction
Layout and Design of the NewsViz software (Figure 4.4) is modelled on the look of similar animation applications such as Flash. This way the experienced animation designer is familiar with the work space. Instead of building an animation himself the NewsViz user only loads, types or pastes a news story into the text editor. Spoken (or written) language is the most natural and easiest input for describing scenes. The user then presses the ‘run’ button and he can watch the visualisation in the preview window. The text processing runs invisibly in the background, but can be monitored in the output window, if the expert mode is chosen. The animation can be (re)played, stopped or rewound. If the user is satisfied he can save/export the animation. If the user prefers to alter the animation manually, he has the options to edit the original text, the processed text or the animation elements. To do so the animation can be stepped through frame by frame on the timeline and a range of basic animation and text editing tools are available. A help link gives instructions on how to use the NewsViz software. The final animations are integrated at the top of the news article’s internet site where they play automatically, but also have controls to stop, replay and rewind. Sound can be turned on or off.
[pic]
Figure 3.4: Interface Layout NewsViz
4 Implementation
The NewsViz prototype (Figure 5.1) demonstrates the process of extracting emotions from natural language input leading to automated animation creation. Four test news articles build the basis for the prototype. Knowledge databases and dictionaries, simulated in XML format, are filled with entries required for exactly these news articles. The part-of-speech tagger Qtag is employed for syntactic analysis enabling the selection of adjectives and nouns and runs as an external program, but its output is directly loaded into the Flash application. The four articles can be opened in the text editor and the options menu offers the selection of the different emotion extraction methods. The language analysis process can be viewed step by step in the expert mode or the animations can be directly loaded into the player. The animation editor provides a timeline to manipulate the initial animation. Interface, graphics and animations of this projects prototype are developed in Macromedia Flash MX and Photoshop. Flash files have the advantage to be small in memory size and quickly viewable through streaming technology. The functionality is programmed in ActionScript. A complete NewsViz animation with manually designed action visualisation is integrated as an swf-file into an HTML news websites (Figure 5.2). The end user needs a Flash Player installed in his browser to be able to view the animations.
[pic] [pic]
Figure 4.1: Interface Screenshot Figure 4.2: Animation Integrated into Website
5 Evaluation and Testing
The NewsViz prototype was tested on a set of four news articles of the same news domain – football match reports. The articles were taken from BBC and FIFA online describing the same two World Cup 2006 matches. The writing style and content structure differs between these two news providers. BBC taking the side of the England team report slightly biased and thus use more emotional expression. FIFA being the official organizer of the tournament try to stay fair and objective towards all teams.
5.1 System Performance
The three different emotion extraction methods, word by word, sentence based and threshold, were run on these news stories with varying word types or word type combinations. The output of NewsViz is evaluated against two forms of human interpretation of the articles. A short manual description outlines the general course of emotion of a match as reported in each article naming three to five emotions. A second more fine grained interpretation assigns one (or two) emotions to each sentence. In correspondence to Beeferman’s probabilistic error metric [Beeferman et al., 1999] three types of emotion extraction error are distinguished. Falsely detected emotions are rated with zero points. Missing emotions were assessed depending on their significance in the text. If the overall feeling of the match was represented, two to three points would be given, but if the main emotions were missing, no points were assigned. Very close, but not exact emotions got a value of four. A correct representation of the course of emotion received five points.
|Method |Word by |Sentence |Threshold | | |total | | | | |Word | |based |2 | |3 | | | |Wordtype | |correct |grain |correct |grain |correct |grain |correct |grain | | |adjectives |3.125 |12 |3.25 |7.5 |2.375 |5 |1.25 |2.3 |2.5 | |nouns | |3.875 |31 |2.625 |9.3 |2.875 |14 |2 |4.8 |2.844 | |both | |4 |33 |2.75 |9.5 |3.5 |18 |1.5 |10 |2.938 | |total | |3.667 |25 |2.875 |8.8 |2.917 |12 |1.583 |5.7 | | |Table 5.1: Total Results Analysis of All Test Texts
The results analysis (Table 5.1) shows that the effectiveness of adjectives or nouns varies from text to text, but generally the best results are achieved with the extraction of both word kinds. In average the word by word method produces emotion sequences with the closest correctness, but unfortunately its output is too fine grained for visualisation. Thirty second long animations are best visualised with two to ten mood swings. This means that some form of summarisation is needed. The integration of a dependency parser, which relates words according to their sentence structure, as well as the extension of the knowledge bases are assumed to bring the desired improvement. Some misinterpretation is also due to false part-of-speech tagging by Qtag which has particular trouble with proper nouns. More accurateness can be achieved through training Qtag on football reports.
Overall the outcome of the NewsViz prototype is satisfactory and proves that news texts are suitable for emotion extraction. The generally different sensations of the two described football matches are distinguishable. Three of the four test texts show good results, but for one article the extracted emotions do not seem to match the human sensation.
5.2 Usability and Acceptance
A second questionnaire gave insight into the end user’s evaluation of the new kind of news presentation. Therefore a first complete, but yet manually designed, news animation was built into the BBC website layout. The overall user reaction was very positive. Most respondents were satisfied with the quality of the animation and confirmed that it represents the emotions evoked by the text. Every viewer desired to see more news animations. Design and loading times were partially criticised, but these problems need to be solved in the action visualisation component of NewsViz, which was beyond the scope of this project.
6 Relation to Other Work
NewsViz is compared to six text-to-visual mapping applications introduced in section 2. Among different multi-modal systems NewsViz is one of the few which uses natural language as input, not a specifically written regular language to create animated output. This is also attempted in the animation systems for the agent ‘Max’ [Becker et al., 2004] and CONFUCIUS [Ma, 2006]. NewsViz implements a similar emotion concept to detect emotional shifts as in Byrne and Max, even though NewsViz aims to solely reflect emotions as they are mentioned in the news articl. To keep the objective and formal character of news reporting, it applies a reduced, universal and ‘personality-free’ version of these concepts for emotion and mood construction. Instead of facial expressions and gestures NewsViz illustrates emotions with basic design principles. NewsViz is the only stystem producing animated 2D output. NewsViz is the first system which offers manual reediting of the outcome.
|multimodal input |language analysis |application area |editable
output |emotion expression |3D |animated | | |text |audio |vision |natural |specific |entertainment |information | |character |universal | | | |WordsEye |X | | | |X |X | | | | |X | | |Mario |X | |X | |X |X | | | | |X |X | |Byrne |X | |X | |X |X |X | |X | |X |X | |Max |X |X |X |X | |X |X | |X | |X |X | |CONFUCIUS |X | | |X | |X | | |X | |X |X | |Picturing Engine |X | |x |X | | |X | | | | | | |SOBA |X | | |X | | |X | | | | | | |NewsViz |X | | |X | | |X |X | |X | |X | |Table 6.1: Comparative Analysis of Text-To-Visual Systems
7 Conclusion and Future Work
The NewsViz prototype extracts sensible words from online football news reports based on an extended dictionary with emotion-indices assigned to each word. The extracted emotions are illustrated in animations. The correctness of the results of this automated process is satisfactory. Technologically NewsViz is viable for the limited topic area and offers a good starting point for more sensible text-to-visual mapping with space for improvement and extension to other topic domains.
The next development steps of the NewsViz emotion extraction tool mainly aim to improve the semantic summarisation of emotions. Football and linguistic knowledge bases are only hinted at in this prototype and need to be fully implemented. SOBA [Buitelaar et al. 2006] offered an efficient technique to automatically generate a football knowledge base. This allows for a better interpretation and representation of new topics and to stay up with current trends and styles. The expanded football knowledge base could set scenes of pre-conditions and context information representing long-term moods. These moods may effect the impact of current event based emotions. A new summarisation method should be tested integrating dependency parser to find related groups of words in sentences, for example an adjective and the noun it specifies, so that the average emotion of these expression can be determined. Falsely tagged part-of-speech elements should be easily reduced through training Qtag. An extended version of WordNet should significantly expand the XML dictionary with the emotion indices and deliver information on related words like synonyms or hyponyms. The synonym sets could be employed for more coherent emotion extraction. Instead of directly loading Flash-animations the output of the emotion extraction tool, should be described in a mark-up language which is compatible with other multi-modal systems, such as EMMA [EMMA, 2006]. For a real-world implementation of the NewsViz system the software and interface logic would be written in C++, Java or similar. SQL databases store, search and update the knowledge data quicker than XML-files.
The development of the event and action visualisation component, as well as the sound and music component were not within the scope of this project. In order to complete the NewsViz software package techniques for automated visualisation need to be considered. This involves either animation of football scenes, players, balls and so on or the selection of appropriate image or video material. WordsEye [Coyne, Sproat, 2001] is an example for object and action modelling, while the Story Picturing Engine[Joschi, Wang, Li, 2004] demonstrates a method to integrate real images. A database of music loops sorted by emotions, a tool to select appropriate music and a speech synthesiser to read out an introduction of the article need to be taken into account for the audio component. A synchronisation component has to bring the background and action animations and the audio together. The functionality to manually edit all these components needs to be implemented. Web crawler software or RSS (Really Simple Syndication) feeds could obtain news articles automatically. The outcome of the emotion extraction tool of NewsViz could also be extended to work in different domains, for example to support decision making in film production for camera work, lighting and perspective.
NewsViz enriches ordinary news websites with attractive and informative animations which is much appreciated by online users overloaded with information. NewsViz brings news reported on the internet closer to the readers, making it more easily understood and memorised. The NewsViz software assists animation designers to cope with the production of daily updated visualisations through creating initial sequences. A full implementation of NewsViz would require a significant amount of work as natural language processing is a research area which is still in its infancy. Nevertheless the NewsViz prototype demonstrates the feasibility and user acceptance of news visualisation as the text analysis processes deliver satisfactory results in emotion extraction. With the suggested improvements higher accuracy can be expected.
References
[Becker et al., 2004] Becker, C., Kopp, S., and Wachsmuth, I. (2004). Simulating the emotion dynamics of a multimodal conversational agent. Affective Dialogue Systems. Springer-Verlag, Berlin: 154-165.
[Beeferman et al., 1999] Beeferman, D., Berger, A., and Lafferty, J. (1999). Statistical models for text segmentation. Machine Learning, Springer Netherlands. (34):177-210.
[Binsted and Luke, 1999] Binsted, K. and Luke, S. (1999). Character Design for Soccer Commentary. Lecture Notes In Computer Science. RoboCup-98: Robot Soccer World Cup II. Springer-Verlag, London: 1604: 22-33.
[Buitelaar et al. 2006] Buitelaar, P., Eigner, T., Gul-rajani, G., Schutz, A., Siegel, M., Weber, N., Cimiano, P., Ladwig, G., Mantel, M., Zhu, H. (2006). Generating and Visualizing a Soccer Knowledge Base. Proceedings of the EACL06 Demo Session, 4/2006:123-126.
[Coyne and Sproat, 2001] Coyne, B. and Sproat, R. (2001). WordsEye: an automatic text-to-scene conversion system. Proceedings of the 28th Annual Conference on Computer Graphics and Interactive Techniques. ACM Press, Los Angeles: 487-496.
[Ekman and Rosenberg, 1997] Ekman, P., Rosenberg E. L., (1997). What the face reveals: Basic and applied studies of spontaneous expression using the facial action coding system. Oxford University Press.
[EMMA, 2006]. and (August, 2006).
[Friedman, 2003] Friedman, D. (2003). Automating Spielberg. Proceedings of the EVA 2003 JML Symposium, London. Invited talk: 1-10.
[Gershon and Page, 2001] Gershon, N. and Page, W. (2001). What Storytelling Can Do for Information Visualisation. Communications of the ACM, 44(8): 31-33.
[Graber, 1990] Graber, D. A. (1990). Seeing Is Remembering: How Visuals Contribute to Learning from Television News.[pic] Journal of Communication. [pic]40(3): 134-55.
[Joschi et al., 2004] Joschi, D. and Wang, J. Z. and Li, J. (2004). The Story Picturing Engine: Finding Elite Images to Illustrate a Story Using Mutual Reinforcement. Proceedings of the 6th ACM SIGMM International Workshop on Multimedia Information Retrieval. ACM Press, New York: 119-126.
[Ma, 2006] Ma, M. (2006). Automatic Conversion of Natural Language to 3D Animation. PhD Thesis, School of Computing and Intelligent Systems, University of Ulster.
[Miller, 1995] Miller, G. A. (1995). WordNet: a lexical database for English. Communications of the ACM, 38(11): 39-41.
[MoodNews, 2005] . (June 2006).
[Qtag, 2003] (July 2006).
[Stokes, 2003] Stokes, N. (2003). Spoken and Written News Story Segmentation Using Lexical Chains. Proceedings of HTL-NAACL 2003, Edmonton, Canada: 49-54.
[TheUnseenVideo, 2005] . (July, 2006).
[UCREL, 2006] . (June, 2006).
................
................
In order to avoid copyright disputes, this page is only a partial summary.
To fulfill the demand for quickly locating and searching documents.
It is intelligent file search solution for home and business.
Related searches
- masters in entrepreneurship salary
- entrepreneurship degree salary
- masters in entrepreneurship jobs
- degree in entrepreneurship online
- masters in entrepreneurship programs
- entrepreneurship programs online
- master s degree in entrepreneurship online
- business administration entrepreneurship jobs
- entrepreneurship degree jobs
- business entrepreneurship degree
- importance of entrepreneurship pdf
- innovation entrepreneurship center