A Survey of Information Retrieval and Recommender System in the era of Large Language Models: Two sides of the same Task

This repository contains the complete dataset, methodology, and analysis for our survey paper:

A Survey of Information Retrieval and Recommender System in the era of Large Language Models: Two sides of the same Task

Our goal is to provide a transparent, reproducible, and lasting resource for the research community. Here you will find the comprehensive list of all papers surveyed, details on our data collection and analysis protocol, and supplemental information that extends the main paper. Since the aim of this project is also to monitor future research publications on the topic, if you believe your work should be included, please feel free to contact any of the corresponding authors.

If you find our survey useful for your research, please consider citing our paper published in Computer Science Review:

@article{Biancofiore2027101058,
title = {A survey of information retrieval and recommender systems in the era of large language models: Two sides of the same task},
journal = {Computer Science Review},
volume = {63},
pages = {101058},
year = {2027},
issn = {1574-0137},
doi = {https://doi.org/10.1016/j.cosrev.2026.101058},
url = {https://www.sciencedirect.com/science/article/pii/S1574013726001668},
author = {Giovanni Maria Biancofiore and Dario {Di Palma} and Claudio Pomo and Ludovico Boratto and Tommaso {Di Noia} and Fedelucio Narducci}
}

Our paper collection followed the PRISMA protocol¹ to ensure a systematic and reproducible search process. The flow of information through the different phases of our review is depicted below.

The four phases of the PRISMA protocol were executed as follows:

1. Identification: An initial set of records was identified through automated searches of major academic databases (e.g., ACM Digital Library, Scopus, Google Scholar) using the following set of keywords:

    "llm recommend",
    "large language model recommend",
    "pre-train recommend",
    "bert recommend",
    "generative recommend",
    "llm information retr",
    "large language model information retr",
    "llm information search",
    "large language model information search",
    "pre-train information retr",
    "pre-train information search",
    "bert information retr",
    "bert information search",
    "generative information retr",
    "generative information search"
   

   The end of this step results into a collection of 2356 scientific papers.
2. Screening: The collected records were first automatically de-duplicated, reaching a total set of 1862 articles. Subsequently, the titles and abstracts of the remaining records were screened against our pre-defined inclusion and exclusion criteria to remove irrelevant studies. Specifically, we excluded works that have not recevied a peer-reviewed process for their publication and those that contains the following excluding keywords in their title:

    "retrieval augmented generation",
    "retrieval-augmented generation",
    "generative adversarial network",
    "GAN",
    "explanation",
    "attack",
    "survey",
    "review"
   

   This phase ends with a refined collection of 1217 papers.
3. Eligibility: The full text of all papers that passed the initial screening was thoroughly reviewed to make a final determination of their eligibility for inclusion in our survey. Furhtermore, we filtered out all those works that have been published within workshop or low-ranked conferences and journals, including only those paper with potential high-impact.
4. Inclusion: The final set of studies that met all criteria was included in our qualitative and quantitative synthesis, with a total of 388 papers. These are the papers presented in the data tables in this repository.

The core of this repository is the complete collection of papers analyzed in our survey. To facilitate exploration and further analysis, we have organized the literature into four distinct tables based on the primary research era and domain:

• Pre-GPT Era: Information Retrieval
• Pre-GPT Era: Recommender Systems
• Modern LLM Era: Information Retrieval
• Modern LLM Era: Recommender Systems

In addition to standard bibliographic data (Title, Authors, Year, Venue), each table includes computed metrics to help gauge the relative impact of each paper over time. These are:

• The Venue Rank: Determined from reliable sources²³, using conference rankings (A*, A, B, etc.) and journal quartiles (Q1, Q2, etc.).
• The Number of Citations: Obtained from Google Scholar, retrieved on January 2025 and updated on February 2026.
• The Normalized Citation Score: A measure designed to normalize the impact of a paper relative to its age. It provides a more balanced view of a paper's influence, especially when comparing newer papers to older, more established ones. It is calculated using the following formula:

  Normalized Citation Score = Citation Score / (Current Year - Publication Year + 1)
  

  Note: The +1 in the denominator is included to avoid division by zero for papers published in the current year and to ensure a fair baseline.

The following tables summarize the collected works, reporting for each entry the title, authors, year, venue, rank, citation count, normalized citation, and taxonomy labels; while all other fields are self-explanatory, the taxonomy labels indicate the categories we defined during our survey, with further details provided in the paper.

This section provides the data behind the trend analysis figures presented in the main paper, along with supplemental information that could not be included due to space constraints.

The figures in our survey that show publication trends, citation velocity, and the distribution of research across subtopics are all generated from the data in this repository, and they are available in the /analysis directory.

Any additional tables, figures, or details that supplement the main manuscript is made available in the /supplement directory.

This page is managed and maintined by:

• Giovanni Maria Biancofiore⁴ giovannimaria.biancofiore@poliba.it
• Dario Di Palma⁴ dario.dipalma@poliba.it
• Claudio Pomo⁴ claudio.pomo@poliba.it
• Ludovico Boratto ludovico.boratto@unica.it
• Tommaso Di Noia tommaso.dinoia@poliba.it
• Fedelucio Narducci fedelucio.narducci@poliba.it