distilbert-no-mans-sky-sentiment

🛰️ DistilBERT — No Man’s Sky Steam Review Sentiment Model

Author: @bkhang
License: CC BY-NC 4.0
Base Model: DistilBERT base uncased
Dataset: Steam Reviews (No Man’s Sky, 2016–2024)

---

📘 Overview

This model is a fine-tuned version of DistilBERT trained on over 300,000 Steam reviews for No Man’s Sky, spanning from 2016 to 2024.
It classifies reviews into two categories:

• 🟩 Recommended → Positive sentiment
• 🟥 Not Recommended → Negative sentiment

The model captures the reputation turnaround of No Man’s Sky — from launch controversies to positive reception after major updates (NEXT, Beyond, Frontiers, etc.).

---

📊 Training Details

Average validation F1-score ≈ 0.98 across time-split samples.

---

🧠 Example Inference

from transformers import AutoTokenizer, AutoModelForSequenceClassification
import torch

model = "bkhang/distilbert-no-mans-sky-sentiment"
tokenizer = AutoTokenizer.from_pretrained(model)
model = AutoModelForSequenceClassification.from_pretrained(model)

review = "After the latest update, it's actually amazing now!"
inputs = tokenizer(review, return_tensors="pt")
outputs = model(**inputs)
probs = torch.softmax(outputs.logits, dim=-1)

labels = ["Not Recommended", "Recommended"]
print(f"{labels[probs.argmax()]} ({probs.max().item():.4f})")