What NLP covers
NLP brings together several levels of working with text — from low-level processing to
understanding meaning:
Basic tasks:
– Tokenization — splitting text into words or word pieces
– Lemmatisation and stemming — reducing to a base form (“bought” → “buy”)
– Text classification — which category a query belongs to
Meaning tasks:
– Named entity recognition (NER) — “Nike Air Max” = brand + model
– Sentiment analysis — is a review positive, neutral or negative
– Intent understanding — does the shopper want to buy, compare or find an offline store
Generation:
– Autosuggest — completing a search query
– Product description generation
– Conversational answers in an AI Shopping Assistant
NLP in e-commerce search
Traditional search is exact or fuzzy string matching. A shopper types “white plimsolls” and the
system looks for those words in product names. If the catalogue says “white casual sneakers”, the
result may simply never be found.
NLP search works through semantic embeddings: the query and the products are converted into vectors
and the search runs on proximity in a meaning space. “White plimsolls” and “casual sneakers white”
are semantically close and return overlapping results.
Important: NLP does not remove the need for a good catalogue. If product cards are poorly
filled in — no attributes, short descriptions — even the best NLP search cannot compensate for the
missing data it has to match against.