Tuesday 11 March 2025
A team of researchers has developed a new system for detecting fake online advertisements, with a focus on real estate listings in Vietnam. The system, which combines machine learning and multimodal processing, has achieved an impressive detection accuracy of 91.5%.
The proliferation of false advertising online is a major concern, as it can lead to financial losses and damage to the reputation of genuine businesses. In the real estate sector, fake ads can be particularly problematic, as they may mislead buyers or renters about properties that do not exist or are not available.
To tackle this issue, the researchers designed an end-to-end system called Fake Advertisements Detection using Automated Multimodal Learning (FADAML). This system uses a combination of natural language processing and computer vision techniques to analyze online real estate listings and identify those that are likely to be fake.
The FADAML system begins by collecting data on genuine and fake ads from popular Vietnamese real estate websites. It then uses machine learning algorithms to extract features from this data, including text-based information such as property descriptions and prices, as well as visual features like images of properties.
Next, the system employs multimodal processing techniques to integrate these different types of features into a single representation. This allows it to capture complex relationships between the various elements of an ad, which can be indicative of its authenticity.
In addition to detecting fake ads, the FADAML system also provides insights into why certain ads are likely to be fraudulent. By analyzing the features that are most commonly associated with fake ads, the system can help real estate professionals and regulators identify common patterns or red flags that may indicate an ad is not genuine.
The researchers tested their system using a dataset of over 29,000 real estate listings in Vietnam, which included both genuine and fake ads. They found that FADAML was able to detect fake ads with an accuracy rate of 91.5%, outperforming several other machine learning models they compared it to.
One of the key advantages of the FADAML system is its ability to handle complex, noisy data – a common challenge in natural language processing and computer vision tasks. By combining multiple modalities and using automated feature extraction, the system can accurately identify fake ads even when the data is incomplete or contains errors.
The potential applications of this technology are significant. In addition to real estate listings, it could be used to detect fake ads in other industries such as e-commerce or employment recruitment.
Cite this article: “Detecting Fake Online Real Estate Listings with Machine Learning and Multimodal Processing”, The Science Archive, 2025.
Fake Advertisements, Online Advertising, Real Estate Listings, Vietnam, Machine Learning, Multimodal Processing, Natural Language Processing, Computer Vision, Data Analysis, Detection Accuracy
