There can be a number of reasons to explain why an index will not generate predictions. Here are some examples:
-
Low number of queries during the training: There is not enough traffic on the index. Query Categorization requires 1000 unique queries over the past 90 days to produce acceptable predictions.
-
Too many facet values: The attribute selected to build the taxonomy has too many facet values, which makes the taxonomy tree too large.
-
Low number of clicks and conversions: If the model doesn't have enough clicks and conversions on queries, it can't accurately predict the categories associated to a query.
-
Low number of unique user tokens: To avoid spambots who could try artificially promoting some items in your index, Algolia uses an aggressive filtering on the events. One of them is based on the user tokens. If, for some reason, the number of user tokens you send are low, the model isn't able to work.
-
Hierarchical attribute not found: One of your attributes wasn't found in the faceting attributes of your record. Consider removing it from the configuration.
-
Hierarchical attribute missing: One of your attributes doesn't seem to appear on most records. Check your indexation process or consider removing it from your configuration.
-
Inconsistent categories' hierarchy tree: Facets set in the categories' hierarchy prevented building a consistent category tree. When using an array of facets, you need to have common prefixes.
-
High number of categories per item: Each item tends to be associated with many categories. This could be due to noisy/redundant facet values. It's recommended you go through your index and remove redundant categories.
-
Category has too many facet values: One of your categories has too many facet values for the model to work. Consider removing it from the configuration.
-
The taxonomy tree is too large: The taxonomy tree generated with the categories you provided is too large. Please cut it down by removing categories from the configuration, or reduce the number of values in these categories.