Bachelor's thesis involving sentiment analysis.

Our project consists of 4 people etc etc.. Marcel will finish the rest of it anyway...

Prerequisites

Type the following command to install required packages :

sudo pip3 install -r requirements.txt

Create following directories to store postproducts of scripts:

mkdir plots

Download the required data sets, unzip them and store the results in data directory.

Test_set_splitter

Component whose job is to split the data into test set and data set. The test set will later be used to validate and rank the analysis approaches used.

It is mandatory to first use:

python3 -m utils.data_exploration -aggregate

To aggregate data in aggregated directory.

Usage

The component consists of a class Test_set_splitter and it's public methods: SplitAll() and SplitAndAppendToExisting()

To use them, the data needs to be in .csv format in folder aggregated. The component will then split the data into test_set/test_set.csv and data_set/data_set.csv

To split all data use:

python3 -m utils.test_set_splitter

To get more verbose information use:

python3 -m utils.test_set_splitter --log

To append new data to existing sets use:

python3 -m utils.test_set_splitter --append [path_to_file]

Data exploration

Data exploration script provides following functions:

• Aggregating data in data folder, which consits of multiple .xml files seperated in each domain. Script then saves data to one .csv file for each domain to aggregated directory

• Preparing results.json file for each domain. File contains results of premilimnary analysis of data sets. Examples of few of the attributes exported to json file:

  • numberOfPositives
  • numberOfNegatives
  • averageTextLengthWhenPolarityPositiveChars
  • averageTextLengthWhenPolarityNegativeChars etc ...

• Prepering plots of selected attributes to visualise the data.

Script can be be started with following arguments:

• -debug - to print error logs
• -dump - to dump the analysis results to results.json file
• -plot - to show plot analysis as well as save plots to plot directory
• -aggregate - to aggregate data in aggregated directory

Known issues:

• 1# at this point results analysis sometimes crashes. This exception is then caught and 21.37 value is saved to results temoporarly

Baseline models

Mean length of text based model

This model uses linear regression, as well as SGD to calculate loss function.

Prerequisites

• have generated test_set and data_set

Usage

To teach model and save it to binary file for later use, type :

python3 -m baselines.mean_length_baseline -teach

To evaluate the model on a test set:

python3 -m baselines.mean_length_baseline -evaluate

Counting text score based on words' polarity

This model calulates border value based on text score which later is used to decide if text has 'positive' or 'negative' polarity. Border value is just mean of all text scores from data_set

Prerequisites

• have generated test_set and data_set

Usage

To teach model and save it to binary file for later use, type :

python3 -m baselines.word_polarity_counting -teach

To evaluate the model on a test set:

python3 -m baselines.word_polarity_counting -evaluate

Aditional options:

--log - to get more information about script process

--parallel - to tech model on multiple processors, by default it uses only one :(

--help - to get more information

Existing Tools

FastText

https://fasttext.cc/

Prerequisites

• have generated test_set and data_set

Usage

If you have not installed FastText just run scrip

fasttext_tool/setup.sh

It will download FastText and install it (one comand requiers sudo privileges).

To test and run FastText just type command

python3 -m fasttext_tool.fast

Script with generated data.train and data.test needed for futher learing and evaluating. If such files exists in fasttext_tool/ then script will skip generaing step. After that script will train model based on data.train and save trained model to model.bin. If such file already exists in fasttext_tool/ then script will skip this step. Last step is evaulating model based on data.train. Results of evaluation will be saved to file fasttext_tool/results.txt and printed on console.