opinion_analyst

Opinion Analyst

Overview

OpinionAnalyst mirrors three interface concepts from 666ghj/BettaFish: sentiment, stance, and bias records. The local specialist computes them from small fixed phrase lists.

The implementation is an offline keyword-and-rule heuristic. It does not run BettaFish models, ingest social or survey data, infer population opinion, or establish a person's beliefs or protected traits.

Prerequisites

• Use Python 3.11+ in a local OSS Agent Lab checkout and run pip install -e ..
• Provide text the user is authorized to analyze and avoid identity or high-stakes profiling.
• Read the runtime contract for heuristic limits and outputs.

Capabilities

• sentiment: Count fixed positive/negative terms and optionally label three known aspects.
• opinion_analysis: Exercise a structured local classification response.
• stance_detection: Match fixed support/opposition phrases, then fall back to sentiment.
• bias_measurement: Count fixed political, emotional, framing, source, and confirmation phrases.

Tools

Tool	Description	Side Effects
analyze_sentiment	Score sentiment (positive/negative/neutral) with confidence and aspects	None
detect_stance	Classify stance (support/oppose/neutral) toward a target	None
measure_bias	Score bias across configurable dimensions, return flags	None

Instructions

1. Choose sentiment, stance, or bias analysis and supply the required text and target/dimensions.
2. Run the local Python API or CLI.
3. Report the returned label as a heuristic signal with its score and matched evidence count.
4. Do not generalize one text sample to a person, group, electorate, or public population.

Examples

Python API

import asyncio
from agents.specialists.opinion_analyst.agent import OpinionAnalystSpecialist
from oss_agent_lab.contracts import Intent, Query, SpecialistRequest

specialist = OpinionAnalystSpecialist()

request = SpecialistRequest(
    intent=Intent(action="sentiment", domain="opinion", confidence=0.9),
    query=Query(user_input="The new policy is an outstanding step forward."),
    specialist_name="opinion_analyst",
)

result = asyncio.run(specialist.execute(request))
print(result.result)
# {'sentiment': {'sentiment': 'positive', 'confidence': 0.65,
#                'aspects': [], 'overall_score': 0.2}}

Stance Detection

request = SpecialistRequest(
    intent=Intent(
        action="stance",
        domain="opinion",
        confidence=0.9,
        parameters={"target": "climate policy"},
    ),
    query=Query(user_input="I strongly support climate policy reforms."),
    specialist_name="opinion_analyst",
    tools_requested=["detect_stance"],
)
result = asyncio.run(specialist.execute(request))

Bias Measurement

request = SpecialistRequest(
    intent=Intent(
        action="bias",
        domain="opinion",
        confidence=0.9,
        parameters={"dimensions": ["political", "emotional"]},
    ),
    query=Query(user_input="The radical left regime is destroying our nation!"),
    specialist_name="opinion_analyst",
    tools_requested=["measure_bias"],
)
result = asyncio.run(specialist.execute(request))

CLI

oss-lab run opinion_analyst "The product quality is excellent and worth every penny."

Output

Sentiment returns a label, confidence, aspects, and score; stance returns support/oppose/neutral plus reasoning; bias returns requested dimension scores and flags. Values reflect built-in lexicons only.

Error Handling

• Empty inputs return neutral/zero-information results; do not interpret them as measured neutrality.
• An unknown bias dimension is scored with the tool's fallback behavior; disclose the limitation.
• Refuse discriminatory profiling, diagnosis, or consequential decisions based on heuristic output.

Resources

Wraps 666ghj/BettaFish. The local code is inspired by the domain but does not embed BettaFish. See the runtime contract.