Selection and ranking of text from highly imperfect transcripts for retrieval of video content

ABSTRACT

In the domain of video content retrieval, we present an approach for selecting words and phrases from highly imperfect automatically generated transcripts. Extracted terms are ranked according to their descriptiveness and presented to the user in a multimedia browser interface. We use sense querying from the WordNet lexical database for our method of text selection and ranking. Evaluation of 679 video summarization tasks from 442 users shows that the method of ranking and emphasizing terms according to descriptiveness results in higher accuracy responses in less time compared to the baseline of no ranking.

REFERENCES

• 1: Campbell, M., Haubold, A., Ebadollahi, S., Naphade, M.R., Natsev, P., Smith, J.R., Tesic, J., and Xie, L. IBM Research TRECVID-2006 Video Retrieval System. Proc. TRECVID 2006 Workshop. NIST Special Publications, 2006.
• 2: Alexander Haubold , John R. Kender, Analysis and Visualization of Index Words from Audio Transcripts of Instructional Videos, Proceedings of the IEEE Sixth International Symposium on Multimedia Software Engineering, p.570-573, December 13-15, 2004
• 3: http://wordnet.princeton.edu
• 4: http://www.aquaphoenix.com/research/vastmm

INDEX TERMS

Primary Classification:
  H. Information Systems
  H.3 INFORMATION STORAGE AND RETRIEVAL
      H.3.3 Information Search and Retrieval
          Subjects: Selection process

Additional Classification:
  H. Information Systems
  H.3 INFORMATION STORAGE AND RETRIEVAL
      H.3.3 Information Search and Retrieval
          Subjects: Information filtering
  H.5 INFORMATION INTERFACES AND PRESENTATION (I.7)
      H.5.1 Multimedia Information Systems
          Subjects: Evaluation/methodology

General Terms:
Design, Experimentation, Measurement, Performance

Keywords:
imperfect transcription, ranking, relevancy, speech transcripts