Skip to Content

Context Setting

Alright, so lets pick an ecommerce site– Say Levi’s-Great Britain http://www.levi.com/GB/en_GB/category/men/collections/levi-collections-whats-new-men

Screen Shot 2016-07-13 at 1.49.00 PM.png

On the top right we see a generic search, that may use Apache SolR. Below we see structured search.

If I wanted to search for “Levis Great Britain Mens new arrival Slim 32″ , I add “Great Britain” only to set the regional context. So let’s analyze different ways in the current system:

1. On the Levi’s site the search “Mens New Arrival Slim 32” does not provide any result

Screen Shot 2016-07-13 at 1.53.53 PM.png

2. On the Levi’s site the search: “Slim 32” provides a lot of data of which a lot is unrelated. For example, “Boot Cut” is displayed whereas I was looking for

Screen Shot 2016-07-13 at 1.54.19 PM.png

3. What if we use Google? Many do this for most online shopping. The top 10 searches provide a lot of contextScreen Shot 2016-07-13 at 1.58.42 PM.png

Screen Shot 2016-07-13 at 1.58.53 PM.png

So, the question is what can be done here? Why utilize Google, when we can implement a change that keeps the user from leaving the site and possibly getting distracted with counter offers.

Proposal

Using Text Analysis in HANA:

  1. Create dictionaries and rules for product filter dimensions and their values
  2. Dictionaries will help identify the filter dimensions from the value typed in. Example: “Slim” would automatically map to “FIT”.
  3. Rules needs to come in action when value does not uniquely identify the filter dimension and we need to do natural language processing. Example: If we have Slim 32, “32” could stand for waist or length. So unless qualified via qualifiers 32 will be used to filter waist and length both. With a rule if we will map “waist” only for sentences like “Slim and Waist 32” or “Slim with 32 waist” etc.
  4. Use the XS API for runtime applying the configuration on the query to bring in the filter route
  5. If the filter route is achieved feed the converted routes from unstructured search to structured search
  6. If the filter route is not achieved as the query was not identified feed the string to free text search of underlying framework

POC Snap Shots:


Scenario 1: Searching on “Slim 32”, results in: “Slim” gets assigned to “FIT”, “32” gets assigned to both “WAIST” and “LENGTH”. Conversion from unstructured to structure search happens.

Screen Shot 2016-07-13 at 2.09.34 PM.png

Screen Shot 2016-07-13 at 2.09.56 PM.png


Scenario 2: Searching on “Originals waist 32″, results in: “Originals” gets normalized to “Original” the base form and maps to “FIT”, “32” gets assigned only to “WAIST”.  Conversion from unstructured to structure search happens

Screen Shot 2016-07-13 at 2.10.32 PM.png

Screen Shot 2016-07-13 at 2.10.25 PM.png


Scenario 3: Searching on “Shirts”, results in a handover to the existing platform search environment as currently this item type is not mapped within Text Analysis in HANA.

Screen Shot 2016-07-13 at 2.10.50 PM.png

Screen Shot 2016-07-13 at 2.11.00 PM.png


Please see the video of POC in action:

To report this post you need to login first.

Be the first to leave a comment

You must be Logged on to comment or reply to a post.

Leave a Reply