Aspect-Based Sentiment Analysis of Corporate HR Survey

Client

The Client is an HR Department of a global US-based corporation. The Client contacted InData Labs, an NLP software development company, to help them process and analyze the results of an annual corporate survey of employees with open text answers to the questions.

Challenge: analyze HR survey data with LLM, make aspect-based survey data sentiment analysis, highlight main problems mentioned by employees

To complete the task, the InData Labs team processed several questions and responses which included open-format answers and level of agreement with a list of statements. Questions covered various aspects of corporate life and employees’ satisfaction allowing for free form responses. There were a total of 11 questions and each of them received over 1400 responses.

The tech challenge was to automatically process all the data, highlight the main problems, define the sentiment in employees’ responses, and summarize their answers. It was essential to provide the HR department with key insights that describe the overall mood of employees in the company and their attitude towards various aspects of corporate life such as job satisfaction, leadership, management, teamwork, salary, etc. The results also needed to be segmented by department, city, country for more detailed analysis.

Solution: aspect-based sentiment analysis of HR survey data, answer summarization, and identification of most common aspects cited by employees

Our team of NLP developers processed the survey results summarizing employees’ sentiments and attitude towards various aspects of corporate life such as leadership, work, communication etc., and analyzed sentiments.

From each employee’s answer we extracted aspects of corporate life they mentioned, and defined emotional tone and sentiments. We grouped these extracted aspects into semantic categories, titled them, and analyzed  the frequency of their mentions in employees’ responses. For answers to each question we’ve conducted sentiment analysis in HR, and visualized the sentiment distribution.

To help the Client understand the employee attitude towards the company, we summarized the employees answers and categorized them, and analyzed sentiment analysis employee engagement. Our data scientists paid attention to the most frequently mentioned aspects and made a table of them in each sentiment group, supporting each aspect with employees’ quotes.

Examples of aspects and categories:

How the final data looks category distribution:

Sentiment distribution within one category:

Approach scheme:

During the project our data scientists worked with multi-language original data (employees’ answers in 5 languages), and open-source LLM (LLAMA3 70b). Our team also used Helsinki to translate final results into English and Python docx to generate the final report.

Take a closer look at the project pipeline:

1. Detect aspects mentioned in answers and give names to those aspects
2. Categorize aspects into several categories
3. Detect sentiment related to these aspects
4. Extract text spans describing the aspect (LLAMA3 70B)
5. Translate all texts into English (Helsinki models)
6. Create summary using different aggregations:
   
   a. For each question for each sentiment value
   
   b. For question in general (LLAMA3 70B)
7. Combine data into docx file for each question and provide
   
   a. Generative summaries
   
   b. Top aspects with statistics
   
   c. Sentiment distribution
   
   d. Example quote for each category/sentiment/question (Python docx).

Result: NLP-powered text analytics for HR

As a result, InData Labs, a natural language processing solutions company, developed an aspect-based sentiment analysis in HR solution for the employees’ survey data analysis. It automatically summarizes all answers to each question and the overall survey data, identifies the most frequently mentioned aspects and provides employees quotes as evidence for the main conclusions made in summary.

We helped the Client to automatically analyze the annual corporate HR survey with open text answers and gain valuable insights into employees’ feelings and emotions. Now it enables the HR team and management to quickly develop strategies to strengthen the company’s HR brand focusing on areas that need improvement.

For more NLP use cases, please visit the page.