determining a plurality of keywords present in at least one of a plurality of encumbered item pages of an electronic commerce (“e-commerce”) channel that have previously been removed from the e-commerce channel as representative of an item that is prohibited from sale through the e-commerce channel;

determining, for each of the plurality of encumbered item pages, a plurality of search terms that were input by a user as part of a query and that resulted in the encumbered item page being returned to the user as responsive to the query;

training, based at least in part on the plurality of keywords and the plurality of search terms, a machine learning model, wherein training includes defining, based at least in part on the plurality of keywords and the plurality of search terms, a plurality of topics;

determining, based at least in part on a keyword included in an item page of an electronic commerce (“e-commerce) channel and a search term included in a query that resulted in the item page being returned as responsive to the query, a total item risk score indicating a likelihood that the item page corresponds to a risky item, wherein the total item risk score is determined based at least in part on the machine learning model; and

in response to determining that the total item risk score exceeds a threshold, removing the item page from the e-commerce channel so that the item corresponding to the item page cannot be purchased through the item page of the e-commerce channel.