Interactive Movie Recommendation Through Latent Semantic Analysis and Storytelling

Kodzo Wegba, Aidong Lu, Yuemeng Li, Wencheng Wang

Abstract

Recommendation systems have become vital components of online services, yet visualization for these systems remains limited. This paper presents an interactive recommendation approach that bridges the gap between complex algorithms and user understanding. We propose a Latent Semantic Model (LSM) that captures statistical features of semantic concepts on 2D domains to abstract user preferences. Furthermore, we introduce an interactive storytelling mechanism to promote communication between the user and the system. Our approach emphasizes interactivity, explicit user input, and semantic information conveyance, making it accessible to general users. Validated with the MovieLens100K dataset, our model demonstrates how latent semantic analysis can enhance personal recommendation through intuitive visualization.

Introduction

Most modern recommendation engines, such as those used by major streaming platforms, operate as "black boxes." While mathematically efficient, they function in high-dimensional latent spaces that are humanly incomprehensible. This creates two primary challenges: a lack of transparency regarding why specific items are recommended and limited user control over the algorithm's direction. This research aims to transform the recommendation process into a transparent, interactive conversation by prioritizing semantic communication and user-steered exploration.

Methodology

Latent Semantic Model (LSM)

The system utilizes Singular Value Decomposition (SVD) on a user-movie rating matrix to generate a latent space. The core innovation lies in the interpretation of this space through "semantic concepts," such as Familiarity vs. Diversity and Typicality vs. Un-typicality.

To make these dimensions useful, the system employs a scoring equation ($D_v$) to identify 2D projections that best separate a user's "liked" and "disliked" regions. By labeling these areas as "Familiar Zones" or "Diverse Zones," the system provides users with a clear mental map of their tastes.

Figure 4: The LSM defines semantic zones such as 'Familiar' and 'Diverse' to help users navigate their preferences based on Like/Dislike clusters.

Storytelling Mechanism

Rather than presenting a static list, recommendations are framed as a "story"—a sequence of movies with a narrative arc. Movies act as "characters," and their roles (e.g., "a typical drama similar to your favorites") are derived from the LSM.

This mechanism follows an interactive loop where users explore data, the system constructs a narrative path through the latent space, and the user provides immediate feedback to refine the story. For example, a "High Diversity" story might begin in a familiar zone and gradually transition toward more unique, highly-rated territory.

Figure 6: The interactive storytelling loop allows users to explore data and refine the narrative through continuous feedback.

Key Insights and Results

  • Recommendation as Narrative: The research suggests that sequential presentation with a logical flow (e.g., "Because you liked X, you might enjoy Y") is more persuasive and understandable than a simple list.
  • User-Centric Dimensions: The system dynamically selects latent dimensions that are locally relevant to an individual's history rather than using a one-size-fits-all global map.
  • Statistical Validation: Analysis of the MovieLens 100K dataset confirmed that the dimension selection process successfully captures highly recommendable movies within the "like" regions of the 2D plots.
  • Efficiency: The preprocessing stage, including SVD and dimension selection, takes only 3–10 seconds for 30,000 ratings, ensuring the system is viable for real-time interaction.

Conclusion and Future Work

While the statistical results are promising, future work should include formal user studies to measure whether the storytelling interface improves discovery compared to traditional lists. Additionally, the current linear narratives could be expanded into branching stories, and the model could be enriched by integrating text-based metadata like movie reviews to further refine the semantic concepts.

Summary updated

Abstract

Recommendation has become one of the most important components of online services for improving sale records, however visualization work for online recommendation is still very limited. This paper presents an interactive recommendation approach with the following two components. First, rating records are the most widely used data for online recommendation, but they are often processed in high-dimensional spaces that can not be easily understood or interacted with. We propose a Latent Semantic Model (LSM) that captures the statistical features of semantic concepts on 2D domains and abstracts user preferences for personal recommendation. Second, we propose an interactive recommendation approach through a storytelling mechanism for promoting the communication between the user and the recommendation system. Our approach emphasizes interactivity, explicit user input, and semantic information convey; thus it can be used by general users without any knowledge of recommendation or visualization algorithms. We validate our model with data statistics and demonstrate our approach with case studies from the MovieLens100K dataset. Our approaches of latent semantic analysis and interactive recommendation can also be extended to other network-based visualization applications, including various online recommendation systems.
BibTeX
@misc{wegba2017interactive,
  author = {Kodzo Wegba and Aidong Lu and Yuemeng Li and Wencheng Wang},
  title = {{Interactive Movie Recommendation Through Latent Semantic Analysis and Storytelling}},
  year = {2017},
  eprint = {1701.00199},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/1701.00199},
}
Figure