TL;DR
This paper tackles the challenge of data quality in art market analysis by proposing a novel framework that combines ontologies, named entity recognition (NER), and machine learning. By enriching auction data with information extracted from text and images, the authors aim to build more robust art market indices and enable deeper quantitative analysis.
What Problem Does This Paper Address?
The art market, despite growing interest as an alternative investment, suffers from significant data quality issues. Sales data from auction houses often contains missing or inconsistent information, hindering reliable quantitative analysis and the construction of accurate market indices. Existing research, while exploring methods like hedonic regression and repeat-sales regression, often overlooks the fundamental problem of data quality. This paper aims to bridge this gap by introducing a comprehensive approach to data enrichment and analysis, making art market data more reliable and actionable.
Methods
The paper introduces a multi-faceted approach:
- Ontology Design: A custom ontology is developed using OWL and RDFS to provide a unified schema for describing art market data, including details about lots, sales, artworks, and artists. This ontology facilitates the integration of diverse data sources.
- Information Extraction: Standard techniques like web crawling, boilerplate removal, and regular expressions are used to extract structured data from auction websites.
- Data Enrichment:
- Named Entity Recognition (NER): NER is applied to text descriptions to identify and extract entities like locations, organizations, and people. These entities are then disambiguated and linked to external knowledge bases like DBpedia.
- Image Analysis (Computer Vision & Machine Learning): To address missing style information, the authors employ a Deep Convolutional Neural Network (CNN) trained on ImageNet. The penultimate layer of the CNN is used as a feature extractor, and a Support Vector Classifier (SVM) with a linear kernel is trained to predict the art movement (style) of a painting.
- Quantitative Analysis: Standard econometric analysis, specifically linear regression, is used to model the relationship between painting prices and their characteristics. This forms the basis for constructing art market indices.
Datasets
- WikiArt.org - Encyclopedia of fine arts: This dataset, containing manually labeled paintings by art movement, was used for training and evaluating the art movement classifiers. The authors selected classes with at least 1,050 images, resulting in 22 classes. 1,000 images were used for training and 50 for testing.
- Christie's Auction Website: Data from a specific Monet painting sale was collected from Christie's website for the case study.
Key Insights and Ideas
- Data Quality is Paramount: The paper strongly emphasizes that the quality of input data is a critical bottleneck in art market analysis, often overlooked in favor of sophisticated modeling techniques.
- Synergy of Techniques: The true power lies in the combination of ontologies for structured data representation, NER for text-based enrichment, and machine learning (especially computer vision) for image-based feature extraction and classification.
- Bridging the Gap: The approach effectively bridges the gap between traditional art market research and cutting-edge computer science techniques, making quantitative analysis more feasible and accurate.
- Automated Style Classification: The ability to automatically classify painting styles (art movements) is a significant contribution, as this information is frequently missing in auction data and crucial for detailed analysis.
Key Results
The paper presents a case study on a single Claude Monet painting.
- Data Enrichment Success: The NER process successfully identified key entities like "London" and "River Thames" from lot descriptions, and "Paul Durand-Ruel" as an important person related to Monet.
- Art Movement Classification: The SVM classifier achieved positive classification scores for various art movements, with "Impressionism" yielding the highest score for the Monet painting. The paper shows a bar chart of classifier scores for different epochs.
- Foundation for Indices: The enriched data, including automatically classified art movements, provides a more comprehensive dataset for building art market indices, allowing for comparisons with other asset classes.
Limitations and Future Work
- Limited Case Study: The current case study is based on a single artwork, which, while illustrative, does not provide broad statistical validation of the entire framework.
- Extensibility: While the ontology is designed to be extensible, the current work focuses specifically on paintings.
- Future Directions:
- Expanding the ontology to include other types of artworks (e.g., sculptures).
- Applying the framework to a larger dataset to build robust art market indices.
- Investigating which variables play the most significant role in describing a lot through statistical analysis.
- Exploring additional data enrichment sources.
- Conducting an extensive study on the behavior and correlation of different indices on an enriched dataset.
- Making data and research publicly available to foster translational efforts in cultural knowledge sharing.
Summary updated