Missing Movie Synergistic Completion across Multiple Isomeric Online Movie Knowledge Libraries

Jiawei Zhang, Chenwei Zhang, Bowen Dong, Yang Yang, Philip S. YuarXiv 2019 · arXiv

Introduction

Online Knowledge Libraries (OKLs) like IMDb and Douban serve as vital repositories for movie enthusiasts and researchers. However, maintaining these databases is a monumental task, often leading to significant gaps where a movie present in one library is missing from another. This paper addresses the Miso-Klic problem (Multiple aligned ISomeric Online Knowledge LIbraries Completion), focusing on "isomeric" libraries—those that describe the same entities but differ in language, cultural context, and user-generated content.

The IDEA Framework

To bridge these gaps, the authors propose IDEA (Integrated Deep alignEd Auto-encoder), a framework that leverages the synergy between disparate libraries to identify and rank missing content. The framework operates in three distinct stages:

1. Movie Embedding

Traditional text matching fails across languages (English vs. Chinese). IDEA utilizes a Deep Auto-encoder to project diverse movie attributes—including titles, storylines, cast, and even the sentiment of user reviews—into a unified numerical vector space. This allows the system to find similarities based on the latent "identity" of a film rather than just its name.

2. Missing Movie Identification

By training on "anchor links" (movies known to exist in both IMDb and Douban), the model learns to align the feature spaces of both libraries. A movie in Library A is flagged as "missing" in Library B if it lacks close neighbors in the shared latent space.

3. Missing Movie Ranking

Because resources for manual entry are limited, IDEA prioritizes missing movies by predicting their potential quality (rating) and popularity (comment count) in the target library. This ensures that high-impact "blockbusters" are added before obscure titles.

Cultural Divergence and Data Characteristics

The study highlights significant differences in how audiences interact with movies across cultures. For instance, Douban users tend to focus heavily on recent releases, whereas IMDb users show a more distributed interest across historical classics and modern films.

Figure 1: Movie distribution by genre and year, illustrating the dense and varied nature of the datasets across different eras.

Furthermore, the distribution of ratings varies significantly by genre and platform, reinforcing the need for a model that understands these "isomeric" differences to predict quality accurately.

Figure 5: Comparison of rating distributions across genres for Douban (top) and IMDb (bottom), showing the variance IDEA must account for when ranking missing movies.

Key Results

The IDEA framework demonstrates remarkable effectiveness compared to traditional methods:

  • Superior Identification: IDEA achieved an F1-score nearly triple that of baseline methods relying on simple title translation and matching (CT-EM/ET-EM).
  • Ranking Precision: The model predicted movie quality and popularity with a Mean Absolute Error (MAE) of less than 0.01, proving that latent features are highly predictive across cultural boundaries.
  • Robustness: Performance remained stable regardless of the size of the "knowledge gap" between the two libraries, indicating the model's reliability in real-world scenarios.

Conclusion

IDEA provides a robust, automated solution for synchronizing global movie databases by treating cross-lingual libraries as isomeric structures rather than simple translation problems. While currently focused on pairs of libraries, future work aims to extend this synergistic completion to multiple libraries simultaneously and address real-time updates as new content is released.

Summary updated

Abstract

Online knowledge libraries refer to the online data warehouses that systematically organize and categorize the knowledge-based information about different kinds of concepts and entities. In the era of big data, the setup of online knowledge libraries is an extremely challenging and laborious task, in terms of efforts, time and expense required in the completion of knowledge entities. Especially nowadays, a large number of new knowledge entities, like movies, are keeping on being produced and coming out at a continuously accelerating speed, which renders the knowledge library setup and completion problem more difficult to resolve manually. In this paper, we will take the online movie knowledge libraries as an example, and study the "Multiple aligned ISomeric Online Knowledge LIbraries Completion problem" (Miso-Klic) problem across multiple online knowledge libraries. Miso-Klic aims at identifying the missing entities for multiple knowledge libraries synergistically and ranking them for editing based on certain ranking criteria. To solve the problem, a thorough investigation of two isomeric online knowledge libraries, Douban and IMDB, have been carried out in this paper. Based on analyses results, a novel deep online knowledge library completion framework "Integrated Deep alignEd Auto-encoder" (IDEA) is introduced to solve the problem. By projecting the entities from multiple isomeric knowledge libraries to a shared feature space, IDEA solves the Miso-Klic problem via three steps: (1) entity feature space unification via embedding, (2) knowledge library fusion based missing entity identification, and (3) missing entity ranking. Extensive experiments done on the real-world online knowledge library dataset have demonstrated the effectiveness of IDEA in addressing the problem.
BibTeX
@misc{zhang2019missing,
  author = {Jiawei Zhang and Chenwei Zhang and Bowen Dong and Yang Yang and Philip S. Yu},
  title = {{Missing Movie Synergistic Completion across Multiple Isomeric Online Movie Knowledge Libraries}},
  year = {2019},
  eprint = {1905.06365},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/1905.06365},
}
Figure