serenakim/solar-sentiment

A fine-tuned RoBERTa model for classifying sentiment about solar energy technologies

More information: https://solarsentiment.org

Model Overview

This model is a RoBERTa-based sentiment classifier fine-tuned on text related to solar energy, rooftop solar adoption, grid benefits, and public attitudes toward solar technologies.

It predicts one of the following classes:

• Positive
• Negative
• Neutral

Load from the Hub

import pandas as pd
import torch
from transformers import AutoTokenizer, AutoModelForSequenceClassification

model_id = "serenakim/solar-sentiment"

# Load tokenizer + model
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForSequenceClassification.from_pretrained(model_id)
model.eval()

Example Usage

# Load CSV
df = pd.read_csv("sample_text.csv") # https://huggingface.co/serenakim/solar-sentiment/blob/main/sample_text.csv
texts = df["text"].astype(str).tolist()

all_preds = []
all_probs = []

for txt in texts:
    enc = tokenizer(
        txt,
        return_tensors="pt",
        truncation=True,
        padding=True,
        max_length=256
    )

    with torch.no_grad():
        out = model(**enc)
        probs = torch.softmax(out.logits, dim=-1)
        pred = probs.argmax(dim=-1).item()

    label = model.config.id2label[pred]

    all_preds.append(label)
    all_probs.append(probs.numpy()[0])

# Add predictions back to DataFrame
df["pred_label"] = all_preds
df["probs"] = all_probs

print(df.head())

Citation

Kim, Serena Y., Crystal Soderman, and Lan Sang. “Sentiment analysis of solar energy in US cities: a 10-year analysis using transformer-based deep learning.” Journal of Computational Social Science, 8(2), 2025. https://link.springer.com/article/10.1007/s42001-025-00365-z