IBM Photo
Ching-Yung Lin -- Projects



[ This page is under construction!! ]



    Social Computing is to study Human/Social Attributes and Dynamics through observable data using multimodality signal understanding techniques. Social Computing aims at (1) modeling and predicting human behavior via computer, and (2) objectively interpreting observations for social science researchers. We shall work closely with social scientists as well as biomedical scientists to achieve these goals.

    In ancient philosophy, there was no difference between studying mathematics, history, poetry or politics. Aristotle studies planetary motion and poetry with the same methods, and Plato mixes geometrical proofs with his demonstration on the state of intrinsic knowledge. Only when the mathematical proof methodology arises over centuries, there becomes a perceived difference between natural scientific fields and humanities fields. Humanities fields suffer from the difficulty of objective and repeatable experiments that arouses a longstanding argument that whether "social science" belongs scientific disciplines. However, similar perceptions can change. For instance, biology -- before Darwin's articles, was resistant to mathematical proofs and models. Its research methodology stayed at descriptive form before Darwin's The Theory of Nature Selection.

   Video understanding is an important media to interpret human behavior. Sociologist would like to see more subjective perceptions being pinned down by more objective measurements. In a Pedagogy study, for instance, researchers would like to understand which teaching method can improve the interaction of teacher-students or how to effectively teach minority students. Usually, they record classroom videos for 3 months to a year and watch them recursively to support their hypotheses. Another example is, at political science study, researchers would like to find out how public consensus is formed through the interaction of people in a meeting. Usually, they need to observe large amount of videos to answer questions like: (1) Will highly educated people show their opinions more than those with lower education?, (2) Will people with lower education modify their thoughts according to other people's opinions more easily than highly educated ones?, (3) Will the final meeting result/conclusion consistent with the opinion of people who dominate the meeting?, and (4) To what level the meeting conclusion will be driven by the shown-up emotion?  In above cases, visual understanding techniques can detect emotions and human interaction types, speech recognition and text understanding can detect the educational level of participants and the influence chains of attendees’ opinions, and face recognition techniques can identify people. These techniques shall objectively convert high volume videos into semantic indices that will be easily transcribed, searched and further interpreted by researchers.

   Using mostly text-based data, such as emails, web access logs, instant messages, publications, etc, is another important way to understand humanity. Krackhardt showed that companies with strong informal networks perform five or six times better than those with weak networks, especially on the long-term performance. Friend and advice networks drive enterprise operations in a way that, if the real organization structure does not match the informal networks, then a company tends to fail. Since Max Weber first studied modern bureaucracy structures in the 1920s, decades of related social scientific researches have been mainly relying on questionnaires and interviews to understand individuals' thoughts and behaviors. However, data collection is time consuming and seldom provides timely, continuous, and dynamic information. With multimodality understanding technologies, we can automatically detect the informal network to suggest structure change for more effective organizational management. 

    Understanding human behavior can help improving our life quality. As the percentage of the elder population continues to rise, advanced health care system has been an active subject of many research projects. For health care applications, there are often critical needs on recognizing a person’s physical location, condition, and activity. Various sensors can be deployed to acquire multimodality data for recognizing a user’s contextual information. With the knowledge of a user’s contextual information, a personal assistant/recommender or warning system can be developed to recommend activities for improving long-term health conditions, raise alarms for emergency situations, and monitor long-term health changes via early-warning detection.

    In addition to providing objective scalable observations to social scientists, we also aim to develop advanced learning machines to model and predict human and society behaviors. For instance, we can allow machines use supervised learning to learn from the experience of psychologists towards predicting the marriage status of couples through interview videos. Text-based analysis can also be used in analyzing the authenticity of arts, e.g., literature, paintings, scientific publications, based on its style, e.g., the author of articles, paintings, etc.

    Understanding human and society behavior is an important emerging research trend. We call this research thread as Social Computing. We see multidisciplinary research is a must towards its success, and computer scientists definitely play a critical role. Recently, we found out that NSF identified Human and Society Dynamics as one of its five priorities, among nanoscience technology, biocomplexisty in environments, mathematical science, and cyberinfrastructure. We are all human, living in societies. We, thus, consider this as a novel field that is worth significant investigation.

Related Projects:

(1) ExpertiseNet   (with Xiaodan Song, Belle L. Tseng and Ming-Ting Sun)

(2) CommunityNet  (with Xiaodan Song, Belle L. Tseng and Ming-Ting Sun)

(3) Community-based Dynamic Recommendation  (with Xiaodan Song, Belle L. Tseng and Ming-Ting Sun)

(4) Information-Flow Driven Personalized Recommendation
(with Xiaodan Song, Belle L. Tseng and Ming-Ting Sun)

(5) Distributed Multimodaliy Sensor System for Sleep Monitoring and Logging   (with Ya-Ti Peng and Ming-Ting Sun)

(6) Multimodality Sensor Network for Human Activity Inferencing  (with Ya-Ti Peng and Ming-Ting Sun)

(7) Understanding Human Interaction in Meetings for Sociological Studies (with Ya-Ti Peng and Dung-Seng Chen)

(8) Developing Smart Mobile Video Sensors for Active Recognition of Human Daily and Social Behaviors (with Victor Sutan and Jason Cardillo)

(9) Establishing Image Quality Indexing Metrics based on Machine Learning of Human Subjective Vision (with Byung Suk Lee)

[ This page is under construbtion!!]

Last Updated: 01/29/2006