Topic Analysis (ABSA)

ABSA (Aspect-Based Sentiment Analysis) extracts specific topics from employee reviews and their associated sentiment.

🎯 What is ABSA?

ABSA goes beyond simple positive/negative classification. It identifies what aspects employees discuss and how they feel about each one.

Example:

"Great salary but terrible management"

Aspect	Sentiment
Salary and Benefits	POSITIVE
Management	NEGATIVE

📊 Predefined Aspects

The system recognizes 10 core workplace topics:

#	Aspect	Keywords
1	Salary and Benefits	pay, bonus, insurance, pension
2	Work-Life Balance	hours, overtime, remote, flexibility
3	Management	boss, leadership, supervision
4	Company Culture	atmosphere, values, team spirit
5	Career Growth	promotion, opportunities, development
6	Job Security	stability, layoffs, restructuring
7	Work Environment	office, facilities, equipment
8	Colleagues	coworkers, team dynamics
9	Training	learning, courses, skills
10	Communication	transparency, feedback, information

🔧 Extraction Methods

1. Rule-Based (Default)

• Fast keyword matching
• ~65% accuracy
• No API costs
• Runs automatically

2. AI-Powered (Gemini)

• Context-aware extraction
• ~85% accuracy
• ~$0.001 per review
• Used for high-priority companies

📈 Dashboard: Topics Tab

The Topics page in the dashboard shows:

Top 10 Discussed Topics

Bar chart of most frequently mentioned aspects with sentiment breakdown.

Topic Insights

• Most positive topic - Highest positive sentiment
• Most negative topic - Highest negative sentiment
• Most discussed - Highest mention count

Detailed Breakdown

Click any topic to see:

• Sample review excerpts
• Sentiment distribution
• Trend over time

🔌 API Access

curl "https://sentryanalytic.com/api/aspects?company=BMW"

Response:

{
  "company": "BMW",
  "aspects": [
    {"aspect": "salary and benefits", "sentiment": "positive", "count": 421},
    {"aspect": "management", "sentiment": "negative", "count": 318},
    {"aspect": "work-life balance", "sentiment": "positive", "count": 256}
  ]
}

🔄 Automation

ABSA runs automatically after scraping:

Scraping Complete → Sentiment Analysis → ABSA → Materialized Views Refresh

No manual intervention needed!

🛠️ Manual Run

For ad-hoc analysis:

cd backend
source venv/bin/activate

# Analyze specific company
python absa_analyzer.py --company "BMW" --limit 200

# Use AI mode (more accurate)
python absa_analyzer.py --company "BMW"

# Rule-based only (faster)
python absa_analyzer.py --company "BMW" --no-ai

📊 Database Schema

CREATE TABLE review_aspects (
    id SERIAL PRIMARY KEY,
    review_id VARCHAR REFERENCES reviews(review_id),
    aspect VARCHAR(100),
    sentiment VARCHAR(20),
    confidence FLOAT,
    snippet TEXT
);

🎨 Use Cases

For Job Seekers

• "What do employees say about work-life balance at BMW?"
• Quickly identify company strengths and weaknesses

For HR Teams

• "Which topics have the most negative feedback?"
• Prioritize improvement areas

For Investors

• "Is there a pattern of management issues?"
• Risk assessment for due diligence

---

ABSA is powered by Gemini 2.5 Flash for AI-powered extraction.