Traits Predictor

Brief Introduction

This is a simple Python personality predictor. Basically, it will cluster users using Big Five test style classification, and try to group users by the same trait to assign trait scores.

Data will not be provided. Please apply your own dataset.

Features

• Predicting users' personality traits.
• Grouping people by users' personality trait

Methods

Data Pre-processing

• Unigram
• Word frequency

Classification

• Stochastic Gradient Descent
• Random Forest
• Multinomial Naive Bayes
• Bernoulli Naive Bayes
• Gradient Boost
• Multi-layer Perceptron

Regression

• Lasso
• SGD
• KNN
• Multi-layer Perceptron
• Gradient Boost

Model Selection

For each trait:

• Classification: model with the highest 5-fold cross score will be selected.
• Regression: model with the lowest 5-fold MSE will be selected.

Data

Validation user selection is randomly selected from here, and the word feature data is collected from here. For the training data, is collected from here.

Files

• fileProcess.py: pulling data from twitter given some test usernames.
• tweetProcess.py: data processing of raw data pulled from fileProcess.py
• featureExtraction.py: feature extraction and I/O to better format for NRC word list.
• trainProcess.py: training data process.
  • Using Machine Learning techniques. This training process is aimed to predict traits' category, i.e., predict if user is an openness person.
  • Using regression method to get score predictions for each trait.
• trainBuild.py: training data pre-process and associated I/O.
• modelRun.py: model application on validation dataset.

Flow of Scripts

Get Validation Data

featureExtraction.py -> fileProcess.py -> tweetProcess.py

Data Train, Test & Validation

trainBuild.py -> modelRun.py

Limitations and Futher Direction

Data and Source

1. There is few golden standard dataset public online, therefore the training dataset in V1 is limited. In futher development, we should look for a good resource of golden standard dataset.
2. In first version, only plain text that can be found in NRC word list is considered as training sample. For futher development, emoticon can be treated as an important feature. For example, XD, :D, ( ͡° ͜ʖ ͡°), (´･ω･`),（　ﾟ Дﾟ）, 😆, 🙄, can help us to understand the emotion behind the tweet greatly. However, some emoticons can be misleading, for example, this one → 🙂. Update: Found this repo on GitHub, listing the polarity on some emoji.
3. Also, all hashtags are removed from original dataset. This processing method could be simple and violent, so in futher development, hashtags can be analyzed as emotion, as mentioned above.

Method

1. In V1, the final score for each trait is calculated using weighted mean; a better evaluation method should be implemented in further development.
2. In V1, single target regression is used to get predicted trait score. In futher development, multiple target regression can be implemented to reduce variance. Moreover, model selection for single target regression can be another way to improvement model. Due to limited training sample, model selection for single target regression is not performed in V1.
3. As mentioned in Data and Source, only plain text that can be found in NRC without Part-of-speech tagging. In futher development, POS can be applied on understanding tweets, to improve classfication and regression performance.

Output

1. Initially, a star graph, such as player attributes in NBA 2K and FM 17 (like this one), is the final output for this project. However I didn't find a suitable tool to visualize the final score. d3.js can be one of the solutions, but it's new to me; needs to take some time to learn it.

Citation

• Nasukawa, T., & Yi, J. (2003, October). Sentiment analysis: Capturing favorability using natural language processing. In Proceedings of the 2nd international conference on Knowledge capture (pp. 70-77). ACM.
• Yang, H., & Li, Y. (2013). Identifying user needs from social media. IBM Research Division, San Jose, 11.
• Gou, L., Zhou, M. X., & Yang, H. (2014, April). KnowMe and ShareMe: understanding automatically discovered personality traits from social media and user sharing preferences. In Proceedings of the 32nd annual ACM conference on Human factors in computing systems (pp. 955-964). ACM.
• Vinciarelli, A., & Mohammadi, G. (2014). A survey of personality computing. IEEE Transactions on Affective Computing, 5(3), 273-291.
• Mohammad, S., Zhu, X., Martin, J.: Semantic role labeling of emotions in tweets. In: Proceedings of the WASSA, pp. 32–41 (2014)
• Farnadi, G., Sitaraman, G., Sushmita, S., Celli, F., Kosinski, M., Stillwell, D., ... & De Cock, M. (2016). Computational personality recognition in social media. User modeling and user-adapted interaction, 26(2-3), 109-142.
• Celli, F., Pianesi, F., Stillwell, D., & Kosinski, M. (2013, June). Workshop on computational personality recognition (shared task). In Proceedings of the Workshop on Computational Personality Recognition.
• Banerjee, N., Chakraborty, D., Dasgupta, K., Joshi, A., Madan, S., Mittal, S., ... & Rai, A. (2009). Contextual analysis of user interests in social media sites-An exploration with micro-blogs. IBM Research Report-RI 09012.