Data and Code from: "Quantum Machine Learning for Near-Term Feasibility: A Survey of Emerging Directions and Challenges"

1. Introduction This repository provides the data and code used in the study “Quantum Machine Learning for Near-Term Feasibility: A Survey of Emerging Directions and Challenges” (preprint available at DOI: http://dx.doi.org/10.2139/ssrn.5613270). The purpose of this package is to ensure transparency and reproducibility of the analyses conducted in the paper, and to support further research on empirical trends and practical feasibility in Quantum Machine Learning (QML). The study investigates potential near-term advantages of QML by conducting a large-scale survey of preprints related to quantum machine learning on arXiv. A total of 2,539 preprints (as of September 2024) were collected, including their metadata, abstracts, and citation counts (retrieved as of October 2, 2024). Two complementary topic extraction approaches were applied: Statistical Topic Extraction — clustering using TF–IDF features and t-SNE to identify topic structures across the full preprint set. Expert Inspection-Based Topic Extraction — manual topic identification based on inspection of the top 10% most-cited preprints (245 papers). For each extracted topic, keyword searches were executed over the entire dataset to compute publication volume and citations, enabling trend analysis over time. Furthermore, for two selected promising application areas — Medicine and Drug Discovery, and Finance and Fraud Detection — manual extraction of use cases was conducted. Then, the use cases were categorized, and the publication volume and number of citations for each category were calculated. This repository includes: the retrieved preprint metadata and processed datasets, covering 2,539 preprints along with citation counts from Semantic Scholar; the code used in four stages of the analysis workflow:(1) collection of preprint metadata and citation information,(2) statistical topic extraction using TF-IDF and t-SNE clustering,(3) extraction of the top 10% most-cited preprints for expert inspection, and(4) keyword-based trend analysis of publication volume and citations; two manually curated files:one containing expert inspection-based topics based on the top 10% most-cited preprints,and another listing use cases for two selected promising application areas. Together, these materials allow reproduction of the analyses presented in the paper and provide a foundation for extended exploration of QML research trends. 2. Repository Structure Overview The repository consists of four analysis modules, a Docker environment for reproducibility, and two manually curated Excel files created through expert inspection.The high-level structure is as follows: /0_docker/ /1_preprints_collection/ /2_topic_extraction/ /3_cited_preprints_extraction/ /4_trend_analysis/ /Expert_Inspection-Based_Topic_Extraction.xlsx /Use_Case_Identification.xlsx Each folder corresponds to one stage of the analysis workflow described in the paper, from data collection to topic extraction and trend analysis. 3. Folder Descriptions /0_docker/ Contains the docker-compose.yml and Dockerfile used to recreate a consistent execution environment for the notebooks. This environment is provided to ensure reproducibility of the analysis pipeline, except for components that depend on external data sources whose contents change over time (e.g., arXiv metadata, citation numbers). /1_preprints_collection/ Contains the notebook 1_preprints_collection.ipynb and its associated outputs. This module corresponds to the data collection stage of the study. The notebook performs: retrieve QML-related preprints from arXiv ("quant-ph" category with “machine learning” in the summary), obtain citation counts from Semantic Scholar, and store the retrieved metadata for downstream analysis. Due to the time-dependent nature of arXiv and Semantic Scholar APIs, the results of this notebook cannot be fully reproduced exactly, but the notebook provides a transparent and verifiable documentation of the data collection procedure. /2_topic_extraction/ Contains 2_topic_extraction.ipynb and its associated inputs and outputs. This module corresponds to the statistical topic extraction approach. The notebook performs: TF–IDF vectorization of abstracts, dimensionality reduction using t-SNE, clustering-based topic extraction across all preprints. Figures generated here use standard Matplotlib settings for reproducibility; publication-ready formatting was applied separately for the paper. /3_cited_preprints_extraction/ Contains 3_cited_preprints_extraction.ipynb and its associated inputs and outputs. This module corresponds to the expert inspection-based topic extraction approach. The notebook performs: extracts the top 10% most-cited preprints (245 papers), prepares the dataset used for manual topic assignment. Figures generated here use standard Matplotlib settings for reproducibility. /4_trend_analysis/ Contains 4_trend_analysis.ipynb andits associated inputs and outputs. This module corresponds to the trend analysis. The notebook performs: keyword-based searches for each topic, aggregation of publication and citation counts over time, trend visualization for each application topic. Figures generated here use standard Matplotlib settings for reproducibility. /Expert_Inspection-Based_Topic_Extraction.xlsx This file contains the results of the expert inspection-based topic extraction. For each of the 245 highly cited preprints, expert annotators assigned: technical topics which refer to specific industrial domains or use cases, and application topics which encompass QML methodologies such as model architectures and performance evaluations, and compiled the list of extracted topics referenced in the preprints. /Use_Case_Identification.xlsx This file contains manually curated use case identification results for the two selected application areas: Medicine and Drug Discovery, and Finance and Fraud Detection. The file includes: extracted use cases, categorical grouping of use cases, publication volume and number of citations for each use case categories.