Twitter Positive Negative Developed by Aaron Acerboni Date started: 21 March 2010

Currently : The basic tally scenario works. The more complex tweet rate is still to be completed.

So searching something obscure which doesn't exceed 100 tweets should prove fruitful.

Most of my testing came about using the keyword "barnacles" :-)

Project purpose

To learn about building open source and redistributable software. To provide software which may be of use to someone.

What is it?

Twitter positive negative is a simple tool for comparing the amount of positivity about a searched twitter subject with it's associated negativity. In theory one should be able to grasp the popularity of a subject.

How does it work? (See Logic)

It utilizes 2 simple PHP objects which allow for data retrieval and data sorting of twitter search's positive and negative semantics.

The Result object may then be used to get the resulted data.

A developer may use this tool to create a visualization of a users search term or this tool may be use to collect data over a period of time.

Planned Features

Simple functionality in 2 objects :

- Fetcher object responsible for data retrieval
    - Allow for different search parameters (search term, a date range etc.)
    - Allow for a more dynamic get data method eg. can request 100 tweets or 1000

- Result object responsible for turning this data into comparison data IE a percentage from a tally
    - Intelligent to grasp popularity of subject and request more pages for a more accurate result
    - Optional method to allow for the retrieval of example tweet based on popularity
    - Optional method to allow for the retrieval of the type of data comparison used

Logic

This tool primarily functions from the TWITTER API's positive and negative sentiment filtering.

Currently, sentiment data is returned from the tool as two percentage values representing positivity and negativity.

These two values are to be derived from two different ways depending on the popularity of a subject.

1. Derived from tally.

For unpopular subjects (100 tweets or lower) a simple tally is done for negative tweets in that subject and for the positive tweets.

2. Derived from 'tweet rate'.

For popular subjects (100 tweets or higher) which span multiple pages, a more complicated approach is taken. The percentage is derived from a 'tweet rate', basically the amount of tweets between the earliest tweet and the latest tweet.

One can expect that positive and negative sentiment will have different tweet rates and a percentage can be made.

This approach is taken simply because we can't request over 1000 tweets from the api and tally that up.

Since twitter limits 100 tweets per page, traversing pages can be an expensive trip. I intend for the Result object to guestimate the page to land on by measuring the time between tweets.

ie. If a subject has tweets seconds apart one can assume its very popular and will have a page 10.

Page guestimation is tricky business because twitter's api does not provide any fallback. If you ask for a page which shouldn't exist then you get an error.

For the sake of development we can hard code a fallback to page=2. Perhaps, in future development this page fallback can be included in the Result object's public interface.

I've also been thinking of combining these two algorithms for a more robust set of data. Really, these two came about because of the limitations of the twitter api in this scenario.