Neural Attentive Session-based Recommendation (1711.04725v1)

Abstract: Given e-commerce scenarios that user profiles are invisible, session-based recommendation is proposed to generate recommendation results from short sessions. Previous work only considers the user's sequential behavior in the current session, whereas the user's main purpose in the current session is not emphasized. In this paper, we propose a novel neural networks framework, i.e., Neural Attentive Recommendation Machine (NARM), to tackle this problem. Specifically, we explore a hybrid encoder with an attention mechanism to model the user's sequential behavior and capture the user's main purpose in the current session, which are combined as a unified session representation later. We then compute the recommendation scores for each candidate item with a bi-linear matching scheme based on this unified session representation. We train NARM by jointly learning the item and session representations as well as their matchings. We carried out extensive experiments on two benchmark datasets. Our experimental results show that NARM outperforms state-of-the-art baselines on both datasets. Furthermore, we also find that NARM achieves a significant improvement on long sessions, which demonstrates its advantages in modeling the user's sequential behavior and main purpose simultaneously.

Neural Attentive Session-based Recommendation

The paper "Neural Attentive Session-based Recommendation" by Jing Li et al. presents a new framework for session-based recommendation in e-commerce scenarios where user profiles are invisible. The authors propose a method to enhance accuracy in recommending items based on user behavior within a single session.

Summary

The proposed framework, named Neural Attentive Recommendation Machine (NARM), addresses the challenge that previous methods overlooked: the user's main purpose in the current session. Most existing methods focus on the sequence of user interactions but fail to emphasize the user's central intent during a session. To remedy this, NARM integrates a hybrid encoder with an attention mechanism that captures both the sequential behavior and the key objectives of the user in a unified session representation.

NARM utilizes a bi-linear matching scheme for scoring candidate items, significantly improving the recommendation accuracy. The framework is trained to jointly learn the representations of items and user sessions and their interactions.

Technical Approach

Global and Local Encoders

NARM combines two encoding mechanisms:

1. Global Encoder: Uses Recurrent Neural Network (RNN) with Gated Recurrent Units (GRU) to model the user's entire sequential behavior.
2. Local Encoder: Incorporates an item-level attention mechanism that identifies and emphasizes the user's main purpose in the session by assigning weights to different clicks.

The unified session representation is formed by concatenating the outputs from both encoders.

Bi-linear Matching Scheme

Instead of using a fully-connected layer as in traditional RNNs, NARM uses a bi-linear similarity function between embeddings of candidate items and the unified session representation. This approach reduces the number of parameters and improves performance.

Experimental Validation

The effectiveness of NARM was validated on two benchmark datasets: YOOCHOOSE and DIGINETICA.

Results:

• NARM outperforms state-of-the-art baselines in terms of Recall@20 and MRR@20.
• Particularly, NARM shows strong performance improvements on longer sessions, demonstrating its efficacy in modeling both sequential behavior and primary user intent.
• On the DIGINETICA dataset, NARM achieved a Recall@20 of 62.58% and MRR@20 of 27.35%.

Implications and Future Directions

NARM's attention mechanism highlights its ability to focus on critical user interactions, thereby accurately capturing user intent. This dual consideration of sequential behavior and specific aims within a session aligns well with the natural browsing patterns observed in users.

Practical Implications:

• Enhanced recommendation accuracy in e-commerce platforms, potentially leading to increased user satisfaction and higher conversion rates.

Theoretical Implications and Future Research:

• The attention mechanism's role could be extended to incorporate more complex user behaviors and interactions across multiple sessions.
• Incorporating additional item attributes (e.g., prices, categories) could further boost recommendation performance.
• Exploring the importance of nearest neighbor sessions could provide new insights into user behavior modeling.

The methodological advancements proposed by NARM set the stage for future developments in session-based recommendation systems, emphasizing the importance of capturing both the sequential patterns and primary user intents within a session.