IssueTracker

Hello,

Thank you for reaching out to us with your request.

We have duly noted your feedback and will thoroughly validate it. While we cannot provide an estimated time of implementation or guarantee the fulfillment of the issue, please be assured that your input is highly valued. Your feedback enables us to enhance our products and services.

We appreciate your continued trust and support in improving our Google Cloud Platform products. In case you want to report a new issue, please do not hesitate to create a new issue on the Issue Tracker providing a detailed description of your issue.

Once again, we sincerely appreciate your valuable feedback; Thank you for your understanding and collaboration.

it can be used but the goal is then to recommend items you might want to buy together with item you are looking at

if you want to show items related to the item being viewed than pdp with ctr or cvr is better.

assigning to our PM to make the final call. but i would strongly advise customer try pdp - results will be better

sorry to calrify, pdp = item view model. we call it pdp internally cause that is the page its expected to be shown on

fbt is not well suited for the case the customer is testing on

Anyone can share the project number?

Hi, this project has 2 FBT models: frequently_frequently_1697420189287 and frequently_frequently_1698697500946. Which is for this bug? Thanks!

hi @li...@google.com , how are you? Please let me know if you have any comments on this bug? were you able to identify improvements?

Thanks

Hi rlombardi@, investigated further and found the root cause for this issue.

The FBT model essentially looks at the user events containing items that are viewed or bought together in the same session. This is the reason behind the recommendations displayed in the two issues that the customer pointed out.

So in the case of the Sylvanian families primeira casa toy item, we found one event where a user actually purchased the pliers along with it: https://screenshot.googleplex.com/3n93mzX6fpEkq8k

And similarly for the case of the TV, there were actually user evens where other TVs were bought together: https://screenshot.googleplex.com/BMKnjK2vm8DgnCV

Hi @di...@google.com , I understand that we are basing in products bought together. But we don't believe that this is happening frequently. So for example in the first case it was probably the only one the case that the person bought the toy together with the pliers. Also if we think about people buying TV it's uncommon frequently buying more than one TV in the same session. So We need to be based on purchases that are really see some frequency of buying together.

Hi @rlombardi, yes. This weird purchase behaviors does not happen frequently. We had an internal discussion for improving the FBT quality on this point, and will have new results soon later. Thanks!

Hi liangtang@google.com , please let me know if there is any improvement or if we have any ETA for it. Thanks

Hi rlombardi@google.com, yes. We are working on the improvement for the FBT model now. We expect to have some internal results next week, but it depends on what results we would have.

Hi liangtang@google.com , do we have any updates on this? Please let me know as they are using their own model as you can see from the screenshot below.

https://screenshot.googleplex.com/AngvGgKUYgMEEK6

Hi @rl...@google.com, For this product, Google FBT's recommendation is this panel? https://screenshot.googleplex.com/BRJsktYVGq3DmKk

for me it's not showing this carrousel. But this should be the Google FBT.

@li...@google.com , please let me know if we made progress? Thanks

@shraygulati, the next A/B test is programmed to start 27th January. We need to have FBT model ready until there.

@li...@google.com , pls let us know if there is any evolution on the FBT model. Thanks

Thanks for the confirm. Will update this bug once we can confirm the model improvement.

@li...@google.com @di...@google.com Do you have any update on this? The customer'll start the next A/B test on 27th January

Yes, we did internal testing for the customer's FBT model and saw the improvements for the problematic queries. Will apply the improvement on the production model. @di...@google.com can comment the current status as well.

@li...@google.com @di...@google.com The customer will start the A/B next Monday (27th January). Is the model ready to start the test?

Hi, we know the next Monday for the A/B testing. We are still tune the FBT model and will give the updates by the end of this week.

Hi cyntiahayama@google.com, rlombardi@google.com The FBT model's prediction quality for infrequent items has been improved now. This resolves the previous issue of irrelevant recommendations and the customer should be able to see relevant results for such products. Here are the results for one of the products that the issue was raised on (pliers being recommended for toy previously) in one of the earlier comments: https://screenshot.googleplex.com/6FHPSPNVGR4R5sE