Visual Sentiment Ontology
Psychological Foundation
Our visual sentiment ontology is built based upon the well-known emotional model called Plutchnik’s Wheel of Emotions [1].
This model includes eight dimensions as:
1. ecstasy → joy → serenity
2. admiration → trust → acceptance
3. terror → fear → apprehension
4. amazement → surprise → distraction 5. grief → sadness → pensiveness
6. loathing → disgust → boredom
7. rage → anger → annoyance
8. vigilance → anticipation → interest
Data-driven Sentiment Word Discovery
For each of the 24 emotions we retrieve images and videos from Flickr and YouTube respectively.
In total we retrieve about 310k images and videos holding about 6M tags. This 6M tags are drawn from a set of 55k distinct tags, each being associated with potentially multiple images or videos.
For tag analysis we remove stop-words and perform stemming on the raw tag meta-data. For each emotion query set we perform tag frequency analysis and rank their top 100 tags.
Finally, the sentiment value of each tag is computed using two popular linguistics based senti- ment models, SentiWordNet [3] and SentiStrength [3].
We retrieve 1146 distinct tags from Flickr and 1,047 distinct tags from YouTube forming the resulting set of 1,771 distinct tags with 1,187 nouns (576 positive and negative one and 611 neutral ones) and 268 positive or negative adjectives.
Adjective Noun Pair Construction
Therefore, we propose adjective nouns combi- nations or Adjective Noun Pairs (ANP) to be the main semantic concept structure of the proposed VSO. Note, that the advantage of ANPs, as compared to nouns or adjectives only, is the capability to turn a neutral noun like “dog” into a strong sentiment ANP like “dangerous dog” by adding an adjective with a strong sentiment like “dangerous”.
We construct and rank ANP candidates, then select only the most frequent and high sentiment value ANPs to form a broad and comprehensive ontology.
Acknowledgements
Please refer to the following paper for more details about the structure and construction procedure of our visual sentiment ontology.
Damian Borth, Rongrong Ji, Thomas Breuel and Shih-Fu Chang. Large-scale Visual Sentiment Ontology and Detectors Using Adjective Noun Pairs. ACM Multimedia 2013 Brave New Idea Track, Under Peer Review.
References
[1] Robert Plutchik. Emotion: A Psychoevolutionary Synthesis. Harper & Row, Publishers, 1980.
[2] A. Esuli and F. Sebastiani. SentiWordnet: A publicly available Lexical Resource for Opinion Mining. LREC, 2006.
[3] M. Thelwall et al. Sentiment Strength Detection in Short Informal Text. J. of the American Soc. for Information Science and Tech., 61(12):2544–2558, 2010.
|