Collaborative filter techniques for generating predicted recommendations in sparse domains
Various embodiments of the present disclosure provide methods, apparatus, systems, computing devices, computing entities, and/or the like for generating predicted recommendations by using an input entity representation, a reference entity representation, and collaborative filtering machine learning model.
1 . A computer-implemented method comprising:
initiating, by one or more processors, the performance of one or more prediction-based actions, wherein the one or more prediction-based actions comprise generating a user interface that provides a predicted recommendation to a client computing device, wherein the predicted recommendation is generated by:
generating an input entity representation based at least in part on one or more entity representation features, wherein the input entity representation comprises an entity representation feature value for each entity representation feature; and
generating, using a collaborative filtering machine learning model, the predicted recommendation based at least in part on the input entity representation, wherein:
(i) the collaborative filtering machine learning model is trained based at least in part on an initialization dataset, wherein:
(a) the initialization dataset is representative of an initial entity-candidate matrix comprising a plurality of rating data fields, and
(b) each rating data field is associated with a unique pair of a reference entity representation and a candidate,
(ii) each data field of a first set of the plurality of rating data fields in the initial entity-candidate matrix comprises an initial rating,
(iii) each data field of a second set of the plurality of rating data fields in the initial entity-candidate matrix comprises a null value,
(iv) the collaborative filtering machine learning model generates a model-predicted rating for each data field of the second set,
(v) a model-predicted entity-candidate matrix is generated, wherein:
(a) the model-predicted entity-candidate matrix comprises the initial rating and the model-predicted rating, and
(b) the model-predicted entity-candidate matrix is represented as a model-predicted dataset, and
(vi) the collaborative filtering machine learning model has been further is trained based at least in part on the model-predicted dataset by:
(a) determining, for the reference entity representation, an entity representation latent feature and a candidate latent feature,
(b) generating, based at least in part on the initial rating,
(1) an entity representation prediction value for the entity representation latent feature, and
(2) a candidate prediction value for the candidate latent feature,
(c) generating, based at least in part on the entity representation prediction value, a reference entity representation embedding for the reference entity representation, and
(d) generating, based at least in part on the candidate prediction value, a candidate embedding for the candidate; and
providing, by the one or more processors and via the user interface, a display of an automated recommendation notification based at least in part on the predicted recommendation.
2 . The computer-implemented method of claim 1 further comprising:
determining a qualifying reference entity representation subset comprising one or more reference entity representations, wherein each reference entity representation of the qualifying reference entity representation subset is associated with a similarity measure satisfying a similarity threshold; and
selecting a reference candidate that corresponds to the reference entity representation of the qualifying reference entity representation subset to include as part of the predicted recommendation.
3 . The computer-implemented method of claim 2 , wherein the candidate is ranked based at least in part on a respective initial rating or a respective model-predicted rating.
4 . The computer-implemented method of claim 1 further comprising:
determining a qualifying reference entity representation subset comprising one or more reference entity representations, wherein each reference entity representation of the qualifying reference entity representation subset is associated with a similarity measure satisfying a similarity threshold; and
selecting the reference entity representation from the qualifying reference entity representation subset to include as part of the predicted recommendation, wherein a respective initial rating or a respective model-predicted rating for the candidate satisfies a candidate rating threshold.
5 . The computer-implemented method of claim 4 , wherein the candidate is ranked based at least in part on the respective initial rating or the respective model-predicted rating.
6 . The computer-implemented method of claim 1 , wherein the model-predicted entity-candidate matrix is generated based at least in part on the reference entity representation embedding for the reference entity representation and the candidate embedding for the candidate.
7 . The computer-implemented method of claim 1 , wherein the initialization dataset is generated from historical data extracted from one or more document data objects.
8 . A system comprising
one or more processors and
one or more non-transitory computer readable media storing processor-executable instructions that, when executed by the one or more processors, cause the one or more processors to perform operations comprising:
initiating the performance of one or more prediction-based actions, wherein the one or more prediction-based actions comprise generating a user interface that provides a predicted recommendation to a client computing device, wherein the predicted recommendation is generated by:
generating an input entity representation based at least in part on one or more entity representation features, wherein the input entity representation comprises an entity representation feature value for each entity representation feature; and
generating, using a collaborative filtering machine learning model, the predicted recommendation based at least in part on the input entity representation, wherein:
(i) the collaborative filtering machine learning model has been is trained based at least in part on an initialization dataset, wherein:
(a) the initialization dataset is representative of an initial entity-candidate matrix comprising a plurality of rating data fields, and
(b) each rating data field is associated with a unique pair of a reference entity representation and a candidate,
(ii) each data field of a first set of the plurality of rating data fields in the initial entity-candidate matrix comprises an initial rating,
(iii) each data field of a second set of the plurality of rating data fields in the initial entity-candidate matrix comprises a null value,
(iv) the collaborative filtering machine learning model generates a model-predicted rating for each data field of the second set,
(v) a model-predicted entity-candidate matrix is generated, wherein:
(a) the model-predicted entity-candidate matrix comprises the initial rating and the model-predicted rating, and
(b) the model-predicted entity-candidate matrix is represented as a model-predicted dataset, and
(vi) the collaborative filtering machine learning model has been further is trained based at least in part on the model-predicted dataset by:
(a) determining, for the reference entity representation, an entity representation latent feature and a candidate latent feature,
(b) generating, based at least in part on the initial rating,
(1) an entity representation prediction value for the entity representation latent feature, and
(2) a candidate prediction value for the candidate latent feature,
(c) generating, based at least in part on the entity representation prediction value, a reference entity representation embedding for the reference entity representation, and
(d) generating, based at least in part on the candidate prediction value, a candidate embedding for the candidate; and
providing, via the user interface, a display of an automated recommendation notification based at least in part on the predicted recommendation.
9 . The system of claim 8 , wherein the operations further comprise:
determining a qualifying reference entity representation subset comprising one or more reference entity representations, wherein each reference entity representation of the qualifying reference entity representation subset is associated with a similarity measure satisfying a similarity threshold; and
selecting a reference candidate that corresponds to the reference entity representation of the qualifying reference entity representation subset to include as part of the predicted recommendation.
10 . The system of claim 9 , wherein the candidate is ranked based at least in part on a respective initial rating or a respective model-predicted rating.
11 . The system of claim 8 , wherein the operations further comprise:
determining a qualifying reference entity representation subset comprising one or more reference entity representations, wherein each reference entity representation of the qualifying reference entity representation subset is associated with a similarity measure satisfying a similarity threshold; and
selecting the reference entity representation from the qualifying reference entity representation subset to include as part of the predicted recommendation, wherein a respective initial rating or a respective model-predicted rating for the candidate satisfies a candidate rating threshold.
12 . The system of claim 11 , wherein the candidate is ranked based at least in part on the respective initial rating or the respective model-predicted rating.
13 . The system of claim 8 , wherein the model-predicted entity-candidate matrix is generated based at least in part on the reference entity representation embedding for the reference entity representation and the candidate embedding for the candidate.
14 . The system of claim 8 , wherein the initialization dataset is generated from historical data extracted from one or more document data objects.
15 . One or more non-transitory computer-readable storage media storing instructions that, when executed by one or more processors, cause the one or more processors to perform operations comprising:
initiating the performance of one or more prediction-based actions, wherein the one or more prediction-based actions comprise generating a user interface that provides a predicted recommendation to a client computing device, wherein the predicted recommendation is generated by:
generating an input entity representation based at least in part on one or more entity representation features, wherein the input entity representation comprises an entity representation feature value for each entity representation feature; and
generating, using a collaborative filtering machine learning model, the predicted recommendation based at least in part on the input entity representation, wherein:
(i) the collaborative filtering machine learning model is trained based at least in part on an initialization dataset, wherein:
(a) the initialization dataset is representative of an initial entity-candidate matrix comprising a plurality of rating data fields, and
(b) each rating data field is associated with a unique pair of a reference entity representation and a candidate,
(ii) each data field of a first set of the plurality of rating data fields in the initial entity-candidate matrix comprises an initial rating,
(iii) each data field of a second set of the plurality of rating data fields in the initial entity-candidate matrix comprises a null value,
(iv) the collaborative filtering machine learning model generates a model-predicted rating for each data field of the second set,
(v) a model-predicted entity-candidate matrix is generated, wherein:
(a) the model-predicted entity-candidate matrix comprises the initial rating and the model-predicted rating, and
(b) the model-predicted entity-candidate matrix is represented as a model-predicted dataset, and
(vi) the collaborative filtering machine learning model is trained based at least in part on the model-predicted dataset by:
(a) determining, for the reference entity representation, an entity representation latent feature and a candidate latent feature,
(b) generating, based at least in part on the initial rating,
(1) an entity representation prediction value for the entity representation latent feature, and
(2) a candidate prediction value for the candidate latent feature,
(c) generating, based at least in part on the entity representation prediction value, a reference entity representation embedding for the reference entity representation, and
(d) generating, based at least in part on the candidate prediction value, a candidate embedding for the candidate; and
providing, via the user interface, a display of an automated recommendation notification based at least in part on the predicted recommendation.
16 . The one or more non-transitory computer-readable storage media of claim 15 wherein the operations further comprise:
determining a qualifying reference entity representation subset comprising one or more reference entity representations, wherein each reference entity representation of the qualifying reference entity representation subset is associated with a similarity measure satisfying a similarity threshold; and
selecting a reference candidate that corresponds to the reference entity representation of the qualifying reference entity representation subset to include as part of the predicted recommendation.
17 . The one or more non-transitory computer-readable storage media of claim 16 , wherein the candidate is ranked based at least in part on a respective initial rating or a respective model-predicted rating.
18 . The one or more non-transitory computer-readable storage media of claim 15 wherein the operations further comprise:
determining a qualifying reference entity representation subset comprising one or more reference entity representations, wherein each reference entity representation of the qualifying reference entity representation subset is associated with a similarity measure satisfying a similarity threshold; and
selecting the reference entity representation from the qualifying reference entity representation subset to include as part of the predicted recommendation, wherein a respective initial rating or a respective model-predicted rating for the candidate satisfies a candidate rating threshold.
19 . The one or more non-transitory computer-readable storage media of claim 18 , wherein the model-predicted entity-candidate matrix is generated based at least in part on the reference entity representation embedding for the reference entity representation and the candidate embedding for the candidate.
20 . The one or more non-transitory computer-readable storage media of claim 15 , wherein the initialization dataset is generated from historical data extracted from one or more document data objects.