Machine learning techniques for generating personalized autocomplete prediction
Various embodiments of the present invention provide methods, apparatus, systems, computing devices, computing entities, and/or the like for performing personalized autocomplete predictions. Certain embodiments of the present invention utilize systems, methods, and computer program products that perform personalized autocomplete predictions using a general search corpus and/or individual curated search corpus.
1 . A computer-implemented method comprising:
receiving, by one or more processors, a search query prefix;
generating, by the one or more processors, a set of session-agnostic autocomplete scores for a set of candidate search results by:
generating a first subset of the set of session-agnostic autocomplete scores for a first subset of the set of candidate search results within an individual curated search corpus based at least in part on an edit distance measure between the search query prefix and each of the first subset of the set of candidate search results, and
generating a second subset of the set of session-agnostic autocomplete scores for a second subset of the set of candidate search results within a general search result corpus based at least in part on a trie-based frequency score that is generated based at least in part on one or more qualifying trie leaf node subsets of a general search result corpus trie data object for the general search result corpus and a general corpus historical frequency score for each of the second subset of the set of candidate search results;
generating, by the one or more processors, a set of session-aware autocomplete scores for the set of candidate search results;
generating (i) a weighted set of session-agnostic autocomplete scores by applying a first weight measure to the set of session-agnostic autocomplete scores and (ii) a weighted set of session-aware autocomplete scores by applying a second weight measure to the set of session-aware autocomplete scores, wherein the first weight measure exceeds the second weight measure; and
wherein (i) a weighted summation, of the set of weighted summations, for the candidate search result comprises a weighted combination of at least one weighted session-agnostic autocomplete score of the weighted set of session-agnostic autocomplete scores and at least one weighted session-aware autocomplete score of the weighted set of session-aware autocomplete scores, and (ii) a weight of the weighted combination is based at least in part on a presence of the candidate search result within the individual curated search corpus or the general search result corpus;
generating, by the one or more processors, a set of hybrid autocomplete scores for the set of candidate search results based at least in part on the set of weighted summations between the first subset of the set of session-agnostic autocomplete scores, the second subset of the set of session-agnostic autocomplete scores, and the set of session-aware autocomplete scores, wherein the set of weighted summations is based at least in part on one or more locations of a candidate search result of the set of candidate search results, and the one or more locations comprise the individual curated search corpus or the general search result corpus;
generating, by the one or more processors, a ranked autocomplete prediction based at least in part on the set of hybrid autocomplete scores; and
initiating, using the one or more processors, the performance of one or more prediction-based actions based at least in part on the ranked autocomplete prediction.
2 . The computer-implemented method of claim 1 , wherein generating a session-aware autocomplete score of the set of session-aware autocomplete scores for the candidate search result of the set of candidate search results comprises:
identifying a search result cluster for the candidate search result;
determining that the search result cluster is in a qualifying search result cluster set for the search query prefix, wherein the qualifying search result cluster set comprises each search result cluster of a set of search result clusters that is generated by performing one or more two-means clustering operations on fixed-size search result representations for the set of candidate search results in a candidate search result corpus;
in response to determining that the search result cluster is in the qualifying search result cluster, generating, using a per-cluster matching machine learning model for the search result cluster, a per-cluster match result indicator for the search result cluster, and
in response to determining that the per-cluster match result indicator for the search result cluster is an affirmative per-cluster match result indicator:
(a) generating, using a cross-cluster ranking machine learning model, a qualifying result ranking score for the candidate search result, and
(b) generating the session-aware autocomplete score for the candidate search result based at least in part on the qualifying result ranking score.
3 . The computer-implemented method of claim 2 , wherein the qualifying search result cluster set comprises each search result cluster that: (i) corresponds to a leaf node of a height-fixed candidate search result corpus trie data object, and (ii) corresponds to the search query prefix.
4 . The computer-implemented method of claim 1 , wherein the individual curated search corpus comprises: (i) one or more visited provider designators for a query-generating user profile associated with the search query prefix, and (ii) for each visited provider designator, one or more related provider designators.
5 . The computer-implemented method of claim 1 , wherein the individual curated search corpus comprises: (i) one or more visited provider designators for a query-generating user profile associated with the search query prefix, and (ii) for each visited provider designator, one or more qualified related provider designators that satisfy one or more provider qualification criteria.
6 . The computer-implemented method of claim 5 , wherein the one or more provider qualification criteria comprise a location criterion characterized by a location feature for the query-generating user profile associated with the search query prefix.
7 . The computer-implemented method of claim 1 , wherein the set of weighted summations is based at least in part on a trie-based frequency score weight measure and an edit distance weight measure.
8 . The computer-implemented method of claim 7 , wherein the edit distance weight measure exceeds the trie-based frequency score weight measure.
9 . The computer-implemented method of claim 1 , wherein the set of session-aware autocomplete scores for the set of candidate search results is based at least in part on a determination that the search query prefix is associated with a series of search queries over a period of time.
10 . A system comprising one or more processors and at least one memory storing processor-executable instructions that, when executed by the one or more processors, cause the one or more processors to perform operations comprising:
receiving a search query prefix;
generating a set of session-agnostic autocomplete scores for a set of candidate search results by:
generating a first subset of the set of session-agnostic autocomplete scores for a first subset of the set of candidate search results within an individual curated search corpus based at least in part on an edit distance measure between the search query prefix and each of the first subset of the set of candidate search results, and
generating a second subset of the set of session-agnostic autocomplete scores for a second subset of the set of candidate search results within a general search result corpus based at least in part on a trie-based frequency score that is generated based at least in part on one or more qualifying trie leaf node subsets of a general search result corpus trie data object for the general search result corpus and a general corpus historical frequency score for each of the second subset of the set of candidate search results;
generating a set of session-aware autocomplete scores for the set of candidate search results;
generating (i) a weighted set of session-agnostic autocomplete scores by applying a first weight measure to the set of session-agnostic autocomplete scores and (ii) a weighted set of session-aware autocomplete scores by applying a second weight measure to the set of session-aware autocomplete scores, wherein the first weight measure exceeds the second weight measure; and
wherein (i) a weighted summation, of the set of weighted summations, for the candidate search result comprises a weighted combination of at least one weighted session-agnostic autocomplete score of the weighted set of session-agnostic autocomplete scores and at least one weighted session-aware autocomplete score of the weighted set of session-aware autocomplete scores, and (ii) a weight of the weighted combination is based at least in part on a presence of the candidate search result within the individual curated search corpus or the general search result corpus;
generating a set of hybrid autocomplete scores for the set of candidate search results based at least in part on the set of weighted summations between the first subset of the set of session-agnostic autocomplete scores, the second subset of the set of session-agnostic autocomplete scores, and the set of session-aware autocomplete scores, wherein the set of weighted summations is based at least in part on one or more locations of a candidate search result of the set of candidate search results, and the one or more locations comprise the individual curated search corpus or the general search result corpus;
generating a ranked autocomplete prediction based at least in part on the set of hybrid autocomplete scores; and
initiating the performance of one or more prediction-based actions based at least in part on the ranked autocomplete prediction.
11 . The system of claim 10 , wherein generating a session-aware autocomplete score of the set of session-aware autocomplete scores for the candidate search result of the set of candidate search results comprises:
identifying a search result cluster for the candidate search result;
determining that the search result cluster is in a qualifying search result cluster set for the search query prefix, wherein the qualifying search result cluster set comprises each search result cluster of a set of search result clusters that is generated by performing one or more two-means clustering operations on fixed-size search result representations for the set of candidate search results in a candidate search result corpus;
in response to determining that the search result cluster is in the qualifying search result cluster, generating, using a per-cluster matching machine learning model for the search result cluster, a per-cluster match result indicator for the search result cluster, and
in response to determining that the per-cluster match result indicator for the search result cluster is an affirmative per-cluster match result indicator:
(a) generating, using a cross-cluster ranking machine learning model, a qualifying result ranking score for the candidate search result, and
(b) generating the session-aware autocomplete score for the candidate search result based at least in part on the qualifying result ranking score.
12 . The system of claim 11 , wherein the qualifying search result cluster set comprises each search result cluster that: (i) corresponds to a leaf node of a height-fixed candidate search result corpus trie data object, and (ii) corresponds to the search query prefix.
13 . The system of claim 10 , wherein the individual curated search corpus comprises: (i) one or more visited provider designators for a query-generating user profile associated with the search query prefix, and (ii) for each visited provider designator, one or more related provider designators.
14 . The system of claim 10 , wherein the individual curated search corpus comprises: (i) one or more visited provider designators for a query-generating user profile associated with the search query prefix, and (ii) for each visited provider designator, one or more qualified related provider designators that satisfy one or more provider qualification criteria.
15 . The system of claim 14 , wherein the one or more provider qualification criteria comprise a location criterion characterized by a location feature for the query-generating user profile associated with the search query prefix.
16 . The system of claim 10 , wherein the set of weighted summations is based at least in part on a trie-based frequency score weight measure and an edit distance weight measure.
17 . The system of claim 16 , wherein the edit distance weight measure exceeds the trie-based frequency score weight measure.
18 . One or more non-transitory computer-readable storage media including instructions that, when executed by one or more processors, cause the one or more processors to perform operations comprising:
receiving a search query prefix; generating a set of session-agnostic autocomplete scores for a set of candidate search results by:
generating a first subset of the set of session-agnostic autocomplete scores for a first subset of the set of candidate search results within an individual curated search corpus based at least in part on an edit distance measure between the search query prefix and each of the first subset of the set of candidate search results, and
generating a second subset of the set of session-agnostic autocomplete scores for a second subset of the set of candidate search results within a general search result corpus based at least in part on a trie-based frequency score that is generated based at least in part on one or more qualifying trie leaf node subsets of a general search result corpus trie data object for the general search result corpus and a general corpus historical frequency score for each of the second subset of the set of candidate search results;
generating a set of session-aware autocomplete scores for the set of candidate search results;
generating (i) a weighted set of session-agnostic autocomplete scores by applying a first weight measure to the set of session-agnostic autocomplete scores and (ii) a weighted set of session-aware autocomplete scores by applying a second weight measure to the set of session-aware autocomplete scores, wherein the first weight measure exceeds the second weight measure; and
wherein (i) a weighted summation, of the set of weighted summations, for the candidate search result comprises a weighted combination of at least one weighted session-agnostic autocomplete score of the weighted set of session-agnostic autocomplete scores and at least one weighted session-aware autocomplete score of the weighted set of session-aware autocomplete scores, and (ii) a weight of the weighted combination is based at least in part on a presence of the candidate search result within the individual curated search corpus or the general search result corpus;
generating a set of hybrid autocomplete scores for the set of candidate search results based at least in part on the set of weighted summations between the first subset of the set of session-agnostic autocomplete scores, the second subset of the set of session-agnostic autocomplete scores, and the set of session-aware autocomplete scores, wherein the set of weighted summations is based at least in part on one or more locations of a candidate search result of the set of candidate search results, and the one or more locations comprise the individual curated search corpus or the general search result corpus;
generating a ranked autocomplete prediction based at least in part on the set of hybrid autocomplete scores; and
initiating the performance of one or more prediction-based actions based at least in part on the ranked autocomplete prediction.