Identifying the Best Rental Comps with Vector Embeddings and ML

One of the most frequent challenges real estate investors face is objectively determining the best comps for potential acquisitions.

In a market with which they are familiar, this may be straightforward – they know the comps because they know the market like the back of their hand. If they want to expand to new markets though, this becomes a bigger challenge – how can they know they are selecting the most relevant comps? Fortunately, advancements in machine learning offer a unique way to solve this problem in any market.

What are Vector Embeddings?

Vector embeddings are a fundamental concept in the field of machine learning, characterized by their remarkable flexibility and practical utility. Widely employed in Natural Language Processing (NLP), recommendation systems, and search algorithms, vector embeddings enable the representation of abstract concepts and objects as lists of numbers, allowing for a wide range of mathematical operations to be performed on them.

One of the most compelling features of vector embeddings is their versatility: they can be employed to represent diverse data types such as audio, video, text, and images. This quality grants vector embeddings a uniquely comprehensive range of applications, making them an essential tool in any data scientist's arsenal.

How Can Vector Embeddings be Used in Real Estate?

Traditionally, comparable properties for homes and apartments have been computed using basic data such as the number of bedrooms, bathrooms, and square footage.  

However, as any seasoned real estate professional can attest, two properties with identical basic features can vary greatly in price. This is where the "feeling" of a property comes into play, which is often conveyed through materials and subtle details that can be difficult to quantify.  

With the rise of machine learning and computer vision, it is now possible to extract a wealth of information from property listing photos, including details such as the quality of materials, lighting, and overall ambiance. This is a ton of extra signal that can be incorporated into property valuations and comparable property detection.  

At HelloData.ai, we have been working on developing advanced machine learning models that measure the similarity across all of those features.  

For example, our models can identify the quality of materials used in a property's construction, such as hardwood floors or high-end appliances, which are often indicators of a higher-end property. They can also recognize features such as natural lighting, facing exposure, and an open floor plan, which are highly sought after by many buyers and can command a premium price.  

While our models are still in the early stages of development, we believe that they have the potential to revolutionize the way that properties are valued and marketed in the real estate business enterprise.

By combining physical attributes like square footage, beds, baths, etc. with the qualitative data from photos (which can now be quantified thanks to vector embeddings!), comps can be automatically identified using all the same information an experienced real estate investor would analyze.

Of course, it's worth noting that this is just one piece of the puzzle when it comes to valuing a property. There are many other factors that need to be taken into account, such as location, market trends, and the overall condition of the property. The problem is, in fact, very similar when it comes to them, and we have already started working on tackling those problems!