Explore

PRD - Interstitial filters

Objective -

Establish high level approach for segregation of different filter classes within HVF and IF to be used for heuristics model setup

Build out the heuristics baseline framework for interstitial filters across core feeds

IF / HVF framework -

Outside IN -

HVFs primarily used for category invariant intent levers like rating / price / new arrivals etc

IFs primarily used to surface feed / category specific levers like brand stores, categories, colour etc

Feature

Amazon

Flipkart

Temu

Myntra

Ajio

Consolidated learnings

HVF

Amazon has HVFs laid out in 2 rows, The first universal HVF pushed toggle for Prime followed by rating, brand and price filters. The next row is dynamic HVFs which are more gender or sub category specific

Flipkart - HVFs only pass in category invariant filters like top brands, new arrivals etc. Similar for Myntra

Temu optimises pricing filters on HVFs followed by rating, category and some category specific attributes

Amazon has each IF pertaining to one filter class label similar as ours, Hierarchy given to Gender followed by category which then extends to size, colour and other taxonomies

Flipkart and Myntra pass on taxonomy filters in IF and also integrate the RE to surface similar search queries with visual references

Temu only surfaces similar search queries interstitially

Ajio pushes brand store cards in interstitial RE

There are no rows in this table

⁠

Labels primarily used across IF / HVF (Meesho) -

HVF usage - Primarily used in category agnostic attributes. Also a function of these mapped in first 4 positions driving the most CTR

⁠

⁠

IF usage - Usage skewed towards category, price and gender due to initial model primarily surfacing these across major REs in the first 3 slots driving most CTR

⁠

⁠

Suggested framework

Utilise HVFs primarily to help channelise immediate intent at start of browsing

These combine high level objectives like -

Low price

Top rated

Promos / Deals / Discounts

Strategic programs

Speed delivery etc

Other highly used feed specific attributes

Utilise IFs to primarily help state intent while browsing -

Taxonomy / category specific attributes (Colour / Fabric / Brands / Size etc)

Category agnostic attributes with multiple options like -

Gender / Category

Price

Rating

Discount

IF model approach -

⁠

⁠

Model building strategy -

Classify a feed as homogenous vs heterogenous basis -

Feed’s name

Feed’s VC distribution

Label prioritisation -

Static slotting Category / Gender as the first IF for heterogenous feeds

For other slots / homogenous feeds -

Compute label priority with a mix of -

Historical filter usage

Most searched attributes for that feed’s primary SSCAT

First build the label and label value prioritisation basis historical filter usage

Check for label and value coverage with this across pareto VC real estates

Build the fallback logic in case coverage is poor i.e. <4 labels and <4 values per label

Step by step details -

Prioritizing label position per real estates (~6-8 labels per feed) -

CLP / collections -

Heterogenous - Force fit category / gender or cat+gender as the first IF

2nd and 3rd slot for these feeds can be prioritised between price, discount and colour

Across homogenous and heterogenous CLPs -

CLPs where filter interaction exists -

Prioritise labels basis historical filter usage (100% for static, 50% for visual) - Only pick labels where unique users applying filter >100

CLPs where filter interaction doesn’t exist -

Prioritise labels basis historical filter usage on similar CLPs of the same SSCAT

Search -

Heterogenous - Force fit category / gender or cat+gender as the first IF

2nd and 3rd slot for these feeds can be prioritised price, discount, combo and colour

Across homogenous and heterogenous CLPs -

Searches where filter interaction exists -

Prioritise labels basis historical filter usage (100% for static, 50% for visual) - Only pick labels where unique users applying filter >100

Searches where filter interaction doesn’t exist -

Prioritise labels basis historical filter usage on similar search query cluster

Prioritising filter label values within labels -

Across REs - We’d be surfacing top X filter label values per feed -

Label value ranking -

Prioritise first the label values which have been filtered most for that RE - Unique user clicks>100

If count of label values through historical filter usage <X then -

Pick label value priority basis the attribute for the pertaining RE (CLP / collection / search) that -

Label values prioritised on the attributes leading to most conversion on the respective real estate

If count of label values through historical usage + RE demand shaping <X

Pick label value priority basis the attribute for the pertaining SSCAT / RE cluster

Label values prioritised on the attributes leading to most conversion on the SSCAT / portfolio

Fallback labels -

Prioritise between category / price / discount / size / colour / combo and fabric to be used as fallback IFs wherever there’s acute data sparsity - Based on platform

Decision points -

Should we keep all label ranking democratic here or prioritize category invariant bits like ratings / price and discount first / have variants to test this?

How should we define heterogenous vs homogenous?

Should we look at any other parameter to define ranking for the base model?

Want to print your doc?
This is not the way.

Try clicking the ⋯ next to your doc name or using a keyboard shortcut (

CtrlP

) instead.