Attention-Based Personalized Encoder-Decoder Model for Local Citation Recommendation.

Area of Science:

• Computer Science
• Information Retrieval
• Artificial Intelligence

Background:

• The exponential growth of scientific literature necessitates efficient methods for literature discovery.
• Existing local citation recommendation systems primarily focus on textual semantic similarity, often overlooking crucial metadata.
• This limitation hinders researchers' ability to find relevant papers effectively.

Purpose of the Study:

• To develop an improved local citation recommendation system that addresses the limitations of existing methods.
• To leverage the encoder-decoder framework, inspired by machine translation, for citation recommendation.
• To enhance recommendation accuracy by incorporating non-textual information such as venue and author data.

Main Methods:

• An attention-based encoder-decoder (AED) model was developed for local citation recommendation.
• The model encodes citation contexts into low-dimensional vectors.
• An attention mechanism, integrating venue and author information, was employed within a recurrent neural network (RNN) decoder to score and rank potential reference papers.

Main Results:

• The proposed AED model demonstrated superior performance compared to three state-of-the-art approaches in experiments.
• Experiments were conducted on the DBLP and ACL Anthology Network (AAN) datasets.
• The integration of venue and author information significantly improved recommendation accuracy.

Conclusions:

• The attention-based encoder-decoder model offers a more effective approach to local citation recommendation.
• Incorporating metadata like venue and author information is crucial for enhancing the precision of citation recommendation systems.
• This research provides a valuable tool for researchers to navigate the growing body of scientific literature more efficiently.