您的瀏覽器不支援JavaScript語法,網站的部份功能在JavaScript沒有啟用的狀態下無法正常使用。

Institute of Information Science, Academia Sinica

Research

Print

Press Ctrl+P to print from browser

2009 Achievements

:::

A Probabilistic Generative Framework for Extractive Broadcast News Speech Summarization

IEEE Transactions on Audio, Speech, and Language Processing 17 (2009): 95-106.

Yi-Ting Chen, Berlin Chen, Hsin-Min Wang

Author Affiliations
  • Institute of Information Science, Academia Sinica

In this paper, we consider extractive summarization of broadcast news speech and propose a unified probabi listic generative framework that combines the sentence generative probability and the sentence prior probability for sentence ranking. Each sentence of a spoken document to be summarized is treated as a probabilistic generative model for predicting the document. Two matching strategies, namely literal term matching and concept matching, are thoroughly investigated. We explore the use of the language model (LM) and the relevance model (RM) for literal term matching, while the sentence topical mixture model (STMM) and the word topical mixture model (WTMM) are used for concept matching. In addition, the lexical and prosodic features, as well as the relevance information of spoken sentences, are properly incorporated for the estimation of the sentence prior probability. An elegant feature of our proposed framework is that both the sentence generative probability and the sentence prior probability can be estimated in an unsupervised manner, without the need for handcrafted document-summary pairs. e experiments were performed on Chinese broadcast news collected in Taiwan, and very encouraging results were obtained.

Extractive spoken document summarization using the probabilistic generative framework

 View Full Paper

A Crowdsourceable QoE Evaluation Framework for Multimedia Content

Proceedings of ACM Multimedia (2009): 491-500.

Sheng-Wei (Kuan-Ta) Chen, Chen-Chi Wu, Yu-Chun Chang, and Chin-Laung Lei

Author Affiliations
  • Institute of Information Science, Academia Sinica

In the design of computer systems, user satisfaction is an important factor (if not the most important factor) in determining the effectiveness of a system's design. In this paper, we consider how to measure user satisfaction systematically and efficiently. Traditionally, researchers have used the Mean Opinion Score (MOS) to measure user satisfaction with multimedia content; that is, experiment participants are asked to rate certain multimedia content with an integer score from 1 to 5. en, the average score is taken as the quality of the content. However, this approach is inecient for three reasons: 1) rating content quality by an absolute score is a dicult process; 2) experiments are expensive in terms of monetary costs (rewards for experiment participants) and labor costs (supervision of participants); and 3) there are limitations in terms of time (it is dicult to hire participants at night), computer hardware, and space. To solve these three problems, we have proposed a novel experiment framework. By using paired comparison and an algorithm to verify the consistency of each participant's input, the proposed framework reduces the diculty of obtaining quality judgments, and enables researchers to invite Internet users to participate in their quality assessment experiments. The evaluation results demonstrate the proposed framework enables researchers to outsource their experiments to an Internet crowd without risking the quality of the results; and, at the same time, obtain a higher level of participant diversity at a lower monetary cost.

Experiment Designs
A comparison of laboratory and crowdsourced experiments in terms of their cost and performance

 View Full Paper

SPATE: Small-group PKI-less Authenticated Trust Establishment

Appeared in Proc. MobiSys 2009 (7th Int'l Conference on Mobile Systems, Applications, and Services, June 22- 25, Wroclaw, Poland), pp. 1-14, best paper award of the conference. To appear in IEEE Transactions on Mobile Computing.

Y.-H. Lin, *A. Studer, H.-C. Hsiao, E. L.-H. Kuo, J. McCune, K.-H. Wang, M. Krohn, P.-L. Lin, A. Perrig, H.-M. Sun, B.-Y. Yang

Author Affiliations
  • Research Center for Information Technology Innovation
  • Institute of Information Science, Academia Sinica

Establishing trust between a group of individuals is a difficult problem. We may simplistically reduce trust establishment to exchange of authenticated public keys. But it remains a very difficult problem. Prior works often assume trusted infrastructure or require an individual to trust unknown entities. The iCAST project by TWISC and CMU's CyLab proposed GAnGS (Gather, Authenticate,'n Group Securely), the first practical and scalable scheme implemented that used physical proximity as a token of trust. However, it provides relatively low probabilistic guarantees of authenticity (95% in some realistic settings).is work also by the iCAST project presents SPATE, a primitive that allows users to establish trust via device mobility and physical interaction. Once the SPATE protocol runs to completion, its participants' mobile devices have authentic data that their applications can use to interact securely (i.e., the probability of a successful attack is 2^{-24}). For this work, we leverage SPATE as part of a larger system to facilitate ecient, secure, and userfriendly collaboration via email, le-sharing, and text messaging services. We have implemented SPATE on iPhones and much less powerful Nokia N70 smartphones. e N70 implementation allows users to establish trust in small groups of up to eight users in less than one minute. e bundled Secure SMS system is a "cool" operation that is easy to operate. Secure SMS and the other example SPATE applications provide increased security with little overhead noticeable to users once keys are established.

Execution Flow of SPATE Exchange
Execution Flow of SPATE Exchange. Step 1: Selection and Counting, Step 2: Commitment, Step 3: Distribution, Step 4: Verification. Steps 1b, 1c and 1d are necessary in our implementation because Bluetooth does not support broadcast.

 View Full Paper