Given your probable customers, the ability to search and easily find the relevant information is the key ingredient in creating a user-friendly website.
User search query and results must answer questions like:
Why was this result included?
How is it relevant to me?
How does it differ from the next result?
– Domain-specific dictionary present
Contextual search works based on anticipating what the individual really wants.
– Individual: user history, profile information
– Demographic profile: age, gender, likely interests
– Interest profile: interest expressed explicitly and implicitly
– Location: country, town, proximity
– Device: iPhone, Android
– Date: weekday/weekend; holidays
– Time: time of the day
– Weather: search for local destination will be impacted by current weather conditions
– Mood: positivity, negativity, excitement, hunger
– Trends in the locality
– Suggest a scope within sorting widget
– Suggest both scope and sorting
– Offer faceted sorting
– Sorting based on our relevance calculations by matching similar customer
conversions (correlation among customers)
– Storing sorting preferences (for example user likes expensive products)
Take into account:
Relevance: if category/scope is evident and returns a high relevance (classifier)
Labelling: sort by : dominant category; e.g. if a user searches for “dine out places near me”;
sort by: price in detected location
Eliminate deals/products with poor return to click investment.
1. Query spectrum
-exact search: handle phonetic misspelling/alternative names : spell check synonym; e.g. “beer cafe/biere cafe”
-product type search: access category; available via faceted search; guide user to relevant category; e.g. “healthy food”
-symptom search: user’s last recourse; make categories problem based; e.g. “live music”
-non product search: help page; company information; e.g. “dominos”
2. Query Qualifiers
-feature search: e.g. “places with open dining space”; all significant attributes Searchable; e.g. features of a place/brand
-thematic search: include fuzzy boundaries/categories;internal tagging of products + query understanding; gifts to be tagged with occasion e.g. “cold weather” -> could indicate jackets, sweaters or deals for hot beverages
-relational search: e.g. movies by a director
-subjective search: e.g. “cheap restaurants”, what constitutes a cheap restaurant according to the user?
3. Query structure
how the query is constructed; linguistic and under what circumstances was it conceived; e.g. “pants” -> “women’s pants”
-abbreviation & symbol: bbq -> barbeque nation or barbeque chicken
-implicit search: user takes search details for granted; check for categorical search + user’s info/history in case of high confidence intent, refine search query -> auto correct/set category
-natural language search: “fashion deals with 20% discount”
-identify each category
-extract feature from dataset; true/false -> boolean feature; tag = 20% is a numerical feature string values are categorical values; gender -> boolean
p(C) = p(C)p(x|C)/p(x)
posterior = prior * likelihood / evidence
e.g. Product data
“title”: “Marks & Spencer”
“meta(features)”: “For him, for her, cashback, discount, location specific details”
Variables affecting product rank:
(assumption: reviews are authentic/semantic analysis of rating handled)
product click, views, conversion, rating, relevance, yesterday’s product performance, product name, description, additional keywords, category, best seller, reviews and rating, product images + User history (purchase history), demographic profile, interest shown, previous engagement with site (likes/comments on a particular type of place/product), own posts/visited places