This is the original version of the ChairsFX Sentimizer app. You can use its logic to build your own custom sentimization tool. It uses open-source machine-learning libraries within a desktop application (GUI not included). There's no limit on how many entries you can sentimize, nor any fees for processing large data sets. This version's workflow:
- The user uploads a CSV file with one set of raw user review data per row.
- Sentimizer will categorize each review as positive, neutral, or negative.
- Sentimizer will plot the positive, neutral, and negative scores into a pie chart.
- Sentimizer will summarize each raw review into a shorter version.
- Once it processes all entries, the user can click 'export'.
- Export will produce a spreadsheet filled with each raw review, a summarized version, and a sentiment score (positive, negative, or neutral)
The finished output that the user can export adds two new columns to the original data set:
- A summarized version of the raw data
- A sentiment score (positive, negative, or neutral).
We ran tests using 200 rows of genuine user reviews. Then, we had a human reviewer manually check each positive, negative, or neutral sentiment score for accuracy. Accuracy is around 90%; the program sometimes confuses neutral and negative sentiments.
This means a human editor should always check the final results for accuracy.
At present, it takes a human reviewer around 3 hours to manually verify 200 rows of Sentimizer output. Version 1.0 aims to cut that down by adding more quick-glance data:
- Categorizing raw review sentences as likes or dislikes.
- Extracting the keywords from all columns of likes and dislike sentences
- Plotting the likes and dislikes as bar charts.
- Exportable charts and CSV data