Mining Opinions in Twitter

NubFinder Research Project

Nub The essence or core of an issue, argument etc. (Wiktionary)

Searching and analyzing user opinions on the Web becomes today a technology very important for modern agile businesses. Opinions can be searched on any subject - a product, an organization, an event, or a topic.  

To achieve these ambitious goals,  NubFinder starts with a more 'simple' task  - classification of emotions in Twitter messages. NubFinder collects Twitter messages that mention  some named entity specified by a user.  Any object with a distinguished name can be a named entity. Examples of named entities are: persons ( Muammar Gaddafi, Barack Obama, etc.), products (Asus Transformer, Android Tablet, Apple iPad, etc.), company names (Grid Dynamics), etc.  Collected messages  are classified by NubFinder according to their content into seven emotional categories: joy, shame, sadness, guilt, fear, disgust and anger.

-  the first incarnation of Twitter Emotion Trend Analyzer and has a similar to NubFinder interface.  Enter query string together with time period in minutes and hit "Analyze" button. NubTrend works on your query for the number of minutes you entered in the "Period" field. NubTrend requests messages from Twitter as soon as you hit "Analyze" button. In response Twitter returns some number of the most recent, related to your query, messages, starting from the most recent one plus some number of the older, back in time messages. On most popular topics, in this first request, Twitter returns no more the 1500 messages. It may return none, if your query is not related to a popular enough topic.
After processing this first bunch of messages, NubTrend continues requesting Twitter for new messages, that Twitter could have possibly received from users since the most recent message in the previous bunch sent to NubTrend. On every succeeding query Twitter returns no more then 100 new messages. (This is how "free Twitter Search API" works)