Kinopoisk Movie Reviews with Sentiment Scores

Encompassing 36,591 movie reviews from the Kinopoisk platform, spanning from July 2004 to November 2012. It includes reviews from the TOP250 and BOTTOM100 rank lists, with sentiment scores labeled as Good, Bad, or Neutral.

Use Cases

• Train a sentiment classification model using the `grade3` labels on 36,591 Russian-language reviews.
• Analyze review trends over the `date` field to study changes in sentiment for movies in the `top250` versus `bottom100` lists.
• Build a text generation model conditioned on the `grade3` sentiment score and `movie_name`.
• Investigate the relationship between `author` activity and the distribution of `grade3` sentiment scores across different movies.

Strengths

• Contains 36,591 labeled reviews, providing a substantial corpus for text analysis.
• Sentiment labels (`grade3`) are provided for all reviews across three categories: Good (27,264), Bad (4,751), and Neutral (4,576).
• Covers a defined 8-year time range from July 2004 to November 2012.
• Data is structured with fields for `movie_name`, `author`, `date`, and `title`, enabling multi-faceted analysis.

Limitations

• The dataset is monolingual (Russian), limiting applicability for models targeting other languages.
• The data is temporally stale, with the latest reviews from 2012, which may not reflect current language or sentiment patterns.
• The source lists (TOP250 and BOTTOM100) may introduce a bias towards extreme opinions, underrepresenting average-rated films.

Provenance

Source

Kinopoisk movie review platform.

Collection Method

Collected from the TOP250 and BOTTOM100 rank lists on Kinopoisk.

Time Range

July 2004 to November 2012.

Freshness

Last updated on the platform in June 2024, but the underlying data covers reviews only up to November 2012.

Geography

Primarily Russia, as the source platform and language (Russian) indicate.