Paper: arXiv 2307.09332

Authors: Christopher Gerling

Abstract

With Company2Vec, the paper proposes a novel application in representation learning. The model analyzes business activities from unstructured company website data using Word2Vec and dimensionality reduction. Company2Vec maintains semantic language structures and thus creates efficient company embeddings in fine-granular industries. These semantic embeddings can be used for various applications in banking. Direct relations between companies and words allow semantic business analytics (e.g. top-n words for a company). Furthermore, industry prediction is presented as a supervised learning application and evaluation method. The vectorized structure of the embeddings allows measuring companies similarities with the cosine distance. Company2Vec hence offers a more fine-grained comparison of companies than the standard industry labels (NACE). This property is relevant for unsupervised learning tasks, such as clustering. An alternative industry segmentation is shown with k-means clustering on the company embeddings. Finally, this paper proposes three algorithms for (1) firm-centric, (2) industry-centric and (3) portfolio-centric peer-firm identification.

Complexity vs Empirical Score

  • Math Complexity: 3.0/10
  • Empirical Rigor: 7.5/10
  • Quadrant: Street Traders — empirical and applied, lower math complexity

Why this score: The paper relies on established, low-complexity methods like Word2Vec and k-means clustering, with minimal advanced mathematical derivations. However, it demonstrates high empirical rigor through a large-scale dataset (42k websites), multiple evaluation metrics (industry prediction, cosine similarity), and concrete banking applications, making it highly backtest-ready.

Research Flowchart

  flowchart TD
  A["Research Goal"] --> B["Data Collection<br>Corporate Websites"]
  B --> C["Word2Vec<br>Word Embeddings"]
  C --> D["Dimensionality Reduction<br>Company Embeddings"]
  D --> E{"Application & Evaluation"}
  E --> F["Industry Prediction<br>Supervised Learning"]
  E --> G["Semantic Analytics<br>Top-N Words"]
  E --> H["Peer Identification<br>Clustering & Algorithms"]
  
  style A fill:#e1f5fe,stroke:#01579b
  style F fill:#f3e5f5,stroke:#4a148c
  style G fill:#e8f5e8,stroke:#1b5e20
  style H fill:#fff3e0,stroke:#e65100

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 introduces Company2Vec, a representation learning model for companies using Word2Vec and dimensionality reduction, and proposes three algorithms for firm‑centric, industry‑centric and portfolio‑centric peer‑firm identification.

Source sentence
Finally, this paper proposes three algorithms for (1) firm‑centric, (2) industry‑centric and (3) portfolio‑centric peer‑firm identification.

Assumes 2

  • Affiliation to a certain industry segment can be used as a proxy for firms' similarities. data
    Source sentence
    Assuming that the affiliation to a certain industry segment can be used as a proxy for firms’ similarities, this data source has proved the most promising results so far.
  • Each company belongs to only one distinct industry sector. model
    Source sentence
    One restricting factor of a classification model lies within the fact that the industry membership of a company is limited to one distinct sector.

Data

  • Company information dataset text-news unknown
    German companies
    Source sentence
    The empirical dataset contains 48,145 German companies.
  • Company webpage data text-news unknown
    German company webpages
    Source sentence
    In this study, a total number of 41,912 unique company webpages were identified and then used for the process of webscraping.

Costs

Not applicable The paper reports no trading or execution result.

Finds 3

  • Using top‑3 predictions, the model classifies over 90% of the companies correctly. predictive-accuracy
    Source sentence
    Using top-3 predictions, the model classifies over 90% of the companies correctly.
  • Logistic regression achieves a top‑1 accuracy of 0.5677 on the imbalanced hold‑out data. predictive-accuracy
    Source sentence
    This baseline probability is outperformed by a top-1 accuracy of 0.5677 of the logistic regression.
  • Industry prediction accuracy improves with more k‑means clusters, converging around k = 400. predictive-accuracy
    Source sentence
    The final number of clusters is set to k = 400, because for both granularity levels the prediction accuracy has somewhat converged at this point.

Checked by others

No other paper in the archive has been found to support or contradict these findings (compared with 15 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.