Dog Talk: Moving towards AI to connect pets to potential adopters

  • Animals
  • Monday, 26 Nov 2018

A cute photo of a rescued animal helps to get adopters. Photo:

There isn’t an ugly dog on the planet, every lopsided ear and snaggle tooth is precious, but writing up an Adopt Me post is a tough gig.

As readers scroll through dozens if not hundreds of posts, rescuers have to hook a click with a super cute photo. They then have to describe the pet’s needs and character in a few pithy and appealing sentences – because readers typically won’t look past the first line.

It’s a job that causes many to quail.

“Everybody wants beautiful, cute and pretty dogs and puppies but the issue is that most of our rescued mixed bred street animals are always not as pretty,” Carnea Lee, Independent Pets Rescuer (IPR), notes practically. “The challenging part is getting adopters to come in and participate to play with the animals, and look at their beauty from within.”

Lee has been rescuing for 14 years and has matched more than 600 pets with owners. “I always try to match pet and owners’ characteristics as closely as possible. For example, an active-lifestyle adopter will enjoy the great outdoors and wish for a companion dog to be part of it.”

“Finding out what adopters need in a pet versus what they want is the trickiest part,” Lynette Lim Chia Ling, an independent rescuer who has rehomed some 250 animals in 13 years, muses. “Puppies especially are just as challenging to look after as babies.

“The majority of adopters, especially those raising families, working full-time, or first-time pet guardians, would actually do much better with mature or senior pets. A good match means less chances of abandonment!”

Jeffery Tan, a private pet rescuer in Langkawi for over 10 years, specialises in rescuing hurt feral stray dogs, a niche that requires special care. “I’ve taken in some 500 dogs over the years, but most are treated and sterilised by the vet, recuperate at my place and then they are released,” Tan shares.

“However, every now and again, one of them will be with me for longer, and become unsuitable for release. I’ve rehomed about 100 of those dogs.”

The challenge is that these animals have been severely traumatised, and need special care.

“Just loving dogs isn’t enough,” Tan observes. “I go to the house, make sure it’s in a safe area, and not in one where the neighbours poison dogs.

“Then once that’s OK, I engage with the adopters to make sure they understand how to work with their pet. Like, for example, a dog who’s been beaten with a broom may panic if you want to sweep. You can work around it easily enough, but it’s a matter of making sure everyone knows how to get along together.”

Humongous challenge

A good pet “adopt me” post therefore has to be cute and appealing, and be designed to match pets and humans in terms of time, activity level, home features, lifestyle, pet knowledge and special needs.

Malaysia’s own has matched 39,000 animals with adopters over the last 10 years, and currently has another 15,000 available for adoption.

This experience means have collected a lot of data about what makes pet adoptions work. However, it’s an artform with rather shaky rules of thumb.

“We want to use Artificial Intelligence (AI) to help shelters and rescuers maximise the attractiveness of their pet profiles,” Andy Koh, co-founder of, says. “The better the profile, the faster and easier it is to get animals adopted. That means more lives saved.”

In order to get the job done, are partnering with Google’s Kaggle, an online community of data scientists and machine learners.

“We know certain things are influential: the pose in the image, buzz words in write-ups, and information such as age, gender and breed,” Koh explains. “With Kaggle’s help, the contestants will build software that will ‘judge’ the profiles. If it all works, then we have a model that people can test new profiles against. It will be a robot advisor.”

Ideally, the software means you can take a picture of the pet, write up a script and submit it for analysis. The results will come back with specific advice, such as “beef up your description about the pet’s activity level” and “that pose is only 4/10 cute”. This should guide rescuers into creating better profiles and give the whole adoption process a shot in the arm.

It’s a hugely ambitious project; however, it is not the first step AI has taken.

“To test the waters, we at have recently developed an AI ‘Cuteness Meter’ that ranks pets’ photos,” Koh says. “The Meter guides rescuers on taking pictures that are more appealing to boost their pet’s exposure and adoption rate. We may integrate this into the AI tools as well once we further refine the system.”

The Kaggle contest will take two to three months. Once that concludes, there is a Phase 2 where winning solutions will be used to develop a suite of AI animal-welfare tools.

“If all goes well, the first release should be slated for mid-2019,” Koh says. “We may also follow up with more fun animal welfare AI competitions on Kaggle.”

If you are a data scientist or AI researcher, and you want to join the competition and improve animal welfare together, stay tuned for PetFinder and Kaggle’s announcement of the competition in early December.

Shelters, rescuers and the general public can help by posting their best to create attractive pet profiles and by promoting active engagement.

“AI systems learn and adapt from best practices,” Koh points out, “so your human effort and knowledge will be put to good use in training our AI tools to be cleverer, which will in turn boost their pet adoption rate.”

In a phrase: the more brains, the better!

Article type: metered
User Type: anonymous web
User Status:
Campaign ID: 7
Cxense type: free
User access status: 3

Across The Star Online

Air Pollutant Index

Highest API Readings

    Select State and Location to view the latest API reading

    Source: Department of Environment, Malaysia