What Do Douban Reviews Reveal? Sentiment Analysis of “The Hunt” TV Drama

Background

On November 6, Hunan TV launched the highly anticipated drama “The Hunt,” which quickly topped viewership charts. However, its Douban rating was only 5.7, far below the 9.1 for “Nirvana in Fire” and 8.3 for “The Disguiser.” The overall rating distribution showed a strong polarization between positive and negative scores.

Rating Distribution

Among 11,463 overall ratings, 1‑star reviews accounted for 28.6% and 5‑star for 25.4%. Positive reviews (4‑5 stars) made up 35.8%, neutral (3 stars) 16.5%, and negative (1‑2 stars) 47.8%.

In contrast, among 5,979 short comments, positive feedback rose to 71%, with only 24% negative. For 392 detailed reviews, 84.7% were positive.

Data Collection

Douban’s anti‑scraping measures require login; after obtaining cookies, the script fetches comments. A sleep interval such as time.sleep(1 + float(random.randint(1, 100)) / 20) is recommended to avoid being blocked.

The crawler retrieved 22,440 popular short comments, as shown in the following screenshots:

Data Cleaning

After collection, the dataset was cleaned to remove duplicates, null entries, and irrelevant fields. The final sample size is illustrated in the following figures:

Sentiment Analysis & Word Cloud

Using SnowNLP, each comment received a sentiment score between 0 and 1; scores above 0.5 indicate positive sentiment, below 0.5 negative. The analysis for November 6‑8 shows a predominance of positive emotions, while the word cloud highlights terms such as “good,” “interesting,” and “expectation.”

Later analysis for November 9‑17 confirms that positive sentiment further outweighs negative, with the word cloud featuring words like “beautiful,” “plot,” “expect,” and “like.”

Conclusion

Despite an overall Douban rating of 6.2 in early November, sentiment analysis of popular short comments and word‑cloud visualization indicate that “The Hunt” is regarded as a sharp, insightful realistic drama that generated strong audience anticipation.

Original code and detailed methodology can be found at https://zhuanlan.zhihu.com/p/31182175