Image-based popularity prediction

A method comprising:

• receiving a search query that includes a text token from a client system;
• generating a set of most frequent text tokens included in titles, item descriptions, or both, of listings of an online marketplace based on a frequency of one or more character strings occurring in the titles, the item descriptions, or both;
• determining that the text token included in the search query matches at least one of the set of most frequent text tokens;
• mapping, by one or more processors, the text token to a first average vector of image features of a plurality of average vectors of image features, the first average vector of image features associated with the at least one of the set of most frequent text tokens:
• determining, by the one or more processors, a plurality of query image relevance scores based on a function of the first average vector of image features and respective image feature vectors for a plurality of images associated with a plurality of items;
• ranking the plurality of items based on the plurality of query image relevance scores;
• and
• causing the client system to display the plurality of items search results in an order in accordance with the ranking.

A non-transitory machine readable storage medium storing instructions, that when executed by one or more processors, causes the one or more processors to perform operations comprising:

• receiving a search query that includes a text token from a client system;
• generating a set of most frequent text tokens included in titles, item descriptions, or both, of listings of an online marketplace based on a frequency of one or more character strings occurring in the titles, the item descriptions, or both;
• determining that the text token included in the search query matches at least one of the set of most frequent text tokens;
• mapping the text token to a first average vector of image features of a plurality of average vectors of image features, the first average vector of image features associated with the at least one of the set of most frequent text tokens;
• determining a plurality of query image relevance scores based on a function of the first average vector of image features and respective image feature vectors for a plurality of images associated with a plurality of items;
• ranking the plurality of items based on the plurality of query image relevance scores;
• and
• causing the client system to display the plurality of items in an order in accordance with the ranking.