Paper: arXiv 2305.13532

Authors: Simerjot Kaur, Andrea Stefanucci, Sameena Shah

Abstract

Determining industry and product/service codes for a company is an important real-world task and is typically very expensive as it involves manual curation of data about the companies. Building an AI agent that can predict these codes automatically can significantly help reduce costs, and eliminate human biases and errors. However, unavailability of labeled datasets as well as the need for high precision results within the financial domain makes this a challenging problem. In this work, we propose a hierarchical multi-class industry code classifier with a targeted multi-label product/service code classifier leveraging advances in unsupervised representation learning techniques. We demonstrate how a high quality industry and product/service code classification system can be built using extremely limited labeled dataset. We evaluate our approach on a dataset of more than 20,000 companies and achieved a classification accuracy of more than 92%. Additionally, we also compared our approach with a dataset of 350 manually labeled product/service codes provided by Subject Matter Experts (SMEs) and obtained an accuracy of more than 96% resulting in real-life adoption within the financial domain.

Complexity vs Empirical Score

  • Math Complexity: 3.0/10
  • Empirical Rigor: 7.0/10
  • Quadrant: Street Traders — empirical and applied, lower math complexity

Why this score: The paper uses relatively straightforward deep learning architectures with minimal advanced mathematical derivation, focusing instead on practical application with real-world financial datasets and reported high classification accuracies.

Research Flowchart

  flowchart TD
  A["Research Goal: Automate Industry & Product/Service Code Classification"] --> B{"Key Challenge: Lack of Labeled Data"};
  B --> C["Methodology: Hierarchical Multi-Class & Multi-Label Classifier"];
  C --> D["Process: Unsupervised Representation Learning"];
  D --> E["Input: 20,000+ Company Dataset"];
  E --> F["Output: Classification System"];
  F --> G["Result 1: 92% Accuracy on 20k Dataset"];
  F --> H["Result 2: 96% Accuracy on 350 SME Labels"];
  G & H --> I["Outcome: High Precision Financial Domain Adoption"];

Evidence Card

What the authors claim, on what data, and who has checked it. Read from the paper's own text by a language model: every quote is verbatim, nothing was re-run. How to read this card

Proposes

The paper proposes a hierarchical multi‑class industry code classifier and a targeted multi‑label product/service code classifier that use unsupervised representation learning.

Source sentence
we propose a hierarchical multi-class industry code classifier with a targeted multi-label product/service code classifier leveraging advances in unsupervised representation learning techniques.

Assumes 4

  • The model can predict codes using only the company’s description as input. data
    Source sentence
    by only using company’s description as input.
  • Limited labeled data (≈20,000 companies) is sufficient for training high‑accuracy classifiers. data
    Source sentence
    We demonstrate how a high quality industry and product/service code classification system can be built using extremely limited labeled dataset.
  • Pitchbook industry codes can be mapped to custom SME codes for 75% of the industry taxonomy. model
    Source sentence
    Through this mapping we were able to generate a labeled dataset containing pitchbook descriptions as inputs and SME’s industry codes as output. However, through this mapping we were only able to generate the labeled dataset for 75% of the industry codes.
  • Unsupervised encoding of code descriptions provides meaningful similarity to company descriptions. model
    Source sentence
    we then used a pre-trained large language model to encode the company descriptions and industry code descriptions into a high dimensional distributed vector representation.

Data

  • Pitchbook company dataset · Pitchbook text-news commercial
    ≈20,000 companies
    Source sentence
    Our model has been developed and tested using licensed data from Pitchbook [11]. The dataset contains details on companies such as a brief description on what the company does as well as pitchbook-defined industry verticals.
  • SME‑labeled validation set text-news proprietary
    350 companies manually labeled by Subject Matter Experts
    Source sentence
    we were provided a small set of 350 companies which were manually labeled by Subject Matter Experts (SMEs) containing the ground truth for both industry and product and service.

Costs

Not applicable The paper reports no trading or execution result.

Finds 2

  • The industry code classifier achieves over 92% classification accuracy on a test set of ~20,000 companies. predictive-accuracy
    Source sentence
    We evaluate our approach on a dataset of more than 20,000 companies and achieved a classification accuracy of more than 92%.
  • Overall, the system obtained more than 96% accuracy when compared with 350 manually labeled product/service codes provided by SMEs. predictive-accuracy
    Source sentence
    we also compared our approach with a dataset of 350 manually labeled product/service codes provided by Subject Matter Experts (SMEs) and obtained an accuracy of more than 96% resulting in real-life adoption within the financial domain.

Checked by others

No other paper in the archive has been found to support or contradict these findings (compared with 8 related papers). Coverage is partial, so this is neither replication nor refutation.

Code

No repository by the authors found in the paper.

Extracted 2026-11-14 · same card as JSON · A language model chose these items: check the source sentence before relying on one.