System and method for machine learning-based delivery tagging
A system including one or more processors; and one or more non-transitory computer-readable media storing computing instructions, that when executed on one or more processors, cause the one or more processors to perform certain operations including: training a first submodel of a machine learning model by at least (i) creating a cumulative addition of light gradient boosting models, and (ii) determining weights for aggregation with probabilities from the light gradient boosting models; generating, using the machine learning model, as trained, classifications for nodes, wherein the classifications comprise unions of outputs of the first submodel of the machine learning model and outputs of a second submodel of the machine learning model; and based on the classifications for the nodes, automatically tagging a portion of the nodes as deliverable in an online platform. Other embodiments are described.
1 . A system comprising:
one or more processors; and
one or more non-transitory computer-readable media storing computing instructions, that when executed on one or more processors, cause the one or more processors to perform operations, the operations comprising:
receiving historical delivery records over a given time period from partners associated with items offered to subregions;
generating nodes for combinations, each node comprising a respective one of the partners, one of the items offered by at least one of the partners, and a respective one of the subregions;
training a first submodel of a machine learning model by at least (i) creating a cumulative addition of light gradient boosting models using first features from the historical delivery records from all partners over a first time period, wherein the first features comprise at least one or more first partner features, one or more first item features, and one or more first region features, and (ii) determining weights for aggregation with probabilities from the light gradient boosting models;
training a second submodel of the machine learning model using second features from historical delivery records from individual partners over a second time period, wherein the second features comprise at least one or more second partner features, one or more second item features, and one or more second region features;
generating, using the machine learning model, as trained, classifications for nodes, wherein the classifications comprise unions of node classification outputs of the first submodel of the machine learning model and node classification outputs of the second submodel of the machine learning model; and
based on the classifications for the nodes, automatically tagging a portion of the nodes as deliverable within a given time window in an online platform.
2 . The system of claim 1 , wherein training the first submodel of the machine learning model further comprises:
training the light gradient boosting models to (i) reduce binary log loss, (ii) increase precision scores, and (iii) increase recall scores.
3 . The system of claim 1 , wherein determining the weights further comprises:
determining the weights using a Bayesian Model Combination at a per-seller level and a full-data level.
4 . The system of claim 1 , wherein generating the classifications further comprises:
generating probability scores for the light gradient boosting models for the nodes.
5 . The system of claim 4 , wherein generating the classifications further comprises:
generating an output of the first submodel by applying the weights, as determined, to the probability scores to generate an aggregate probability.
6 . The system of claim 1 , wherein
training the second submodel of the machine learning model further comprises using numerical features from historical delivery records across the second time period and a third time period.
7 . The system of claim 6 , wherein:
training the second submodel further comprises:
training one or more CatBoost models using the numerical features.
8 . The system of claim 7 , wherein:
training the second submodel further comprises:
determining one or more thresholds for the one or more CatBoost models based on second probability scores for the nodes; and
generating the classifications comprises:
generating outputs of the second submodel.
9 . The system of claim 1 , wherein the operations further comprise:
monitoring on-time-delivery (OTD) performances of the nodes over a subsequent time period after the portion of the nodes was tagged as deliverable in the given time window;
automatically un-tagging a first node of the nodes when an OTD performance for the first node falls below one or more un-tagging thresholds; and
automatically re-tagging a second node of the nodes as deliverable in the given time window when an OTD performance for the second node exceeds one or more re-tagging thresholds.
10 . A method being implemented via execution of computing instructions configured to run on one or more processors and stored at one or more non-transitory computer-readable media, the method comprising:
receiving historical delivery records over a given time period from partners associated with items offered to subregions;
generating nodes for combinations, each node comprising a respective one of the partners, one of the items offered by at least one of the partners, and a respective one of the subregions;
training a first submodel of a machine learning model by at least (i) creating a cumulative addition of light gradient boosting models using first features from the historical delivery records from all partners over a first time period, wherein the first features comprise at least one or more first partner features, one or more first item features, and one or more first region features, and (ii) determining weights for aggregation with probabilities from the light gradient boosting models;
training a second submodel of the machine learning model using second features from historical delivery records from individual partners over a second time period, wherein the second features comprise at least one or more second partner features, one or more second item features, and one or more second region features;
generating, using the machine learning model, as trained, classifications for nodes, wherein the classifications comprise unions of node classification outputs of the first submodel of the machine learning model and node classification outputs of the second submodel of the machine learning model; and
based on the classifications for the nodes, automatically tagging a portion of the nodes as deliverable within a given time window in an online platform.
11 . The method of claim 10 , wherein training the first submodel of the machine learning model further comprises:
training the light gradient boosting models to (i) reduce binary log loss, (ii) increase precision scores, and (iii) increase recall scores.
12 . The method of claim 10 , wherein determining the weights further comprises:
determining the weights using a Bayesian Model Combination at a per-seller level and a full-data level.
13 . The method of claim 10 , wherein generating the classifications further comprises:
generating probability scores for the light gradient boosting models for the nodes.
14 . The method of claim 13 , wherein generating the classifications further comprises:
generating an output of the first submodel by applying the weights, as determined, to the probability scores to generate an aggregate probability.
15 . The method of claim 10 , wherein
training the second submodel of the machine learning model further comprises using numerical features from historical delivery records across the second time period and a third time period.
16 . The method of claim 15 , wherein:
training the second submodel further comprises:
training one or more CatBoost models using the numerical features.
17 . The method of claim 16 , wherein:
training the second submodel further comprises:
determining one or more thresholds for the one or more CatBoost models based on second probability scores for the nodes; and
generating the classifications comprises:
generating outputs of the second submodel.
18 . The method of claim 10 , further comprising:
monitoring on-time-delivery (OTD) performances of the nodes over a subsequent time period after the portion of the nodes was tagged as deliverable in the given time window;
automatically un-tagging a first node of the nodes when an OTD performance for the first node falls below one or more un-tagging thresholds; and
automatically re-tagging a second node of the nodes as deliverable in the given time window when an OTD performance for the second node exceeds one or more re-tagging thresholds.
19 . The method of claim 10 , wherein the training the first submodel of the machine learning model further comprises training, using the one or more first partner features as inputs, a first one of the light gradient boosting models to output a first probability score.
20 . The method of claim 19 , wherein the training the first submodel of the machine learning model further comprises:
training, using the first probability score of the first one of the light gradient boosting models, the one or more first partner features, and the one or more first item features as inputs, a second one of the light gradient boosting models, to output a second probability score; and
aggregating the first and the second probability scores using Bayesian Model Combination.