| Model | P | R | F1 | |---------------------------|--------|--------|--------| | RAKE | 0.42 | 0.35 | 0.38 | | mBERT NER | 0.65 | 0.58 | 0.61 | | YAKE (multi) | 0.51 | 0.48 | 0.49 | | Proposed Hybrid | 0.76 | 0.72 | 0.74 |
The hybrid model significantly improves recall by correctly identifying multi-word Hinglish keyphrases (e.g., "superhit picture", "time waste movie").
The extraction and analysis reveal a growing interest in accessible, categorized movie databases. For viewers interested in Hindi and English cinema, these platforms offer a convenient way to explore content. However, challenges such as content rights, regional limitations, and user preferences continue to pose challenges. extraction2020720phindienglishvegamoviesn hot
This report interprets the query string as likely referencing a media file or dataset related to the film "Extraction" (release identifiers), a date (2020-07-20), sources or languages (ph = possibly "ph" for Philippines or "phindi" indicating "ph" + "indienglish" → Hindi/English), a platform or site (vegamovies / vegamoviesn), and a tag "hot". The goal below is to provide structured findings, risks, and recommended next steps for safe, legal, and effective handling.
The exponential growth of user-generated content on streaming platforms and social media has led to a surge in code-mixed text, particularly Hindi-English (Hinglish). Extracting meaningful keyphrases from such unstructured data remains challenging due to lexical variations, lack of standardized grammar, and resource scarcity. This paper proposes a hybrid keyphrase extraction model combining statistical features (TF-IDF, TextRank) with a lightweight neural sequence labeler. Evaluated on a manually annotated corpus of 5,000 movie review sentences from online forums, the proposed model achieves an F1-score of 0.74, outperforming baseline methods by 12%. The approach demonstrates robust performance on named entities, movie titles, and sentiment-bearing phrases. | Model | P | R | F1
Traditional methods for keyphrase extraction include:
Recent work on Hinglish (Kumar et al., 2022) highlights the need for language-agnostic statistical signals combined with contextual embeddings. Recent work on Hinglish (Kumar et al
Given a code-mixed Hindi-English sentence ( S = w_1, w_2, ..., w_n ), the goal is to extract a set of keyphrases ( K = k_1, k_2, ..., k_m ) where each ( k_j ) is a contiguous subsequence of ( S ) representing a salient concept. Keyphrases can be single words (unigrams) or multi-word expressions (up to 3 grams).
Challenges specific to Hinglish: