Friends, we know many of you have been anxiously awaiting our update on Nestoria Rank (our proprietary algorythm for determining relevance – first discussed here) and today your patience is rewarded.
The Nestoria team has been hard at work since our last update. We've added new ways to filter the property listings, new ways to use the map to narrow your property search and loads of new types of local information to qualify an area (photos, government data, more transport data, and council tax data).
All of that is great, but it doesn't really address the issue of relevance. As anyone who works in the search industry will atest, even measuring relevancy is a non-trivial challenge. When results are relevant, you don't even notice, you just have a great search experience. So we thought we'd let you know about some of the challenges we've been grappling with.
- Comprehensiveness: We've significantly grown the size and geographic coverage of our database. Whether you're looking to buy a bungalow in Belfast, or a two bedrooms in the North Laines of Brighton, you can find it on Nestoria.
- Natural Language Processing (NLP): Unfortunately much of the data we're dealing with is unformatted text manually created by humans. We love humans generally, but unfortunately humans make spelling errors, typos, are sloppy with cut and paste, etc. So we've developed some algorithms to parse human entered text and extract the true meaning that's relevant to the home searcher. For example, does a property have a garden or a balcony, or how many bathrooms the house has. You can expect a lot more progress on this front shortly.
- Disambiguation of place names: Relevancy in property search has two main pieces: determining the correct location for a query, and then ordering the results in a relevant way. Disambiguation is a key part of the first part of relevancy. In the UK there are many place names that exist multiple times. For example there are several towns named "Newcastle". Building logic to know when to default to a certain location (in this case Newcastle upon Tyne and not Newcastle, Staffordshire) and when not to default (for example is someone searches for "Bangor") isn't simple.
- Ranking based on quality of a listing: We've begun experimenting with a very involved quality score for each listing based on many factors like how well we were able to geocode the listing, age of the listing, quality of the description (as determined by NLP), and a variety of other factors. This is another area we'll be focusing on heavily in the coming weeks.
Of course we have a few other little secrets as well. We can't reveal everything here, but we can reveal that we remain hard at work on relevancy. As always, it's an ongoing challenge, and one we're excited about diving deeper and deeper into.
Have an opinion about how we're doing? We'd love to hear from you.
to this: 
