Only available this morning: Oct 23 2009. We anticipate many issues. Let us know if you see any. Your comments are most welcome.
Introduction
Crowdmood is an on-going research project on sentiment analysis (or opinion mining) based on a sample of the Twitter data.
Instruction: You can search/add up to 5 classes of entities
(products or brands) and compare opinions or buzz on them. The entity classes are separated by commas.
Each entity class is represented by one or more keywords/keyphrases
separated by "OR". For example, if you want to compare opinions or buzz
of Wii, Xbox and Playstation 3, you can either input in the search box,
"wii, xbox, ps3 OR ps 3 OR playstation 3" (no double quotes), or add
each entity class one by one by inputting and clicking the add button
(Note: there are different ways to say Playstation 3).
You can also delete an entity class by clicking on the cross sign.
Sentiment chart: X - date, Y - percentage of positive opinions.
Buzz chart: X - date, Y - frequency count (number of mentions)
Data: The data is streamed using the Twitter API, which gives
a sample of about 2 million tweets a day. Out of these, most are in
English. In this project, we use only English tweets. We have a text
classifier to remove non-English ones. It should be noted that we do
not do search at the Twitter site using your query. So whatever you
see is based on the random sample of the Twitter data that we received.
Update: The indexed data is updated hourly. It takes a few minutes to update. During this time, it will be slow as it is running on the same machine (we only have one machine for this).
Lost data: Since we have only one machine and we are using the school's network, we lose data when one of them is down. This has already happened a few times.
We are trying to develop a model that will allow people to submit some competing entities and track their sentiments and buzz long term and also see the relevant tweets. That may not be free :-).