System and method for hierarchical category classification of products
This disclosure relates generally to system and method for hierarchical category classification of products. Generally in supervised hierarchical classification, the hierarchy structure is predefined. However, majority of the current machine learning methods either expect the model to learn the hierarchy from the data or requires separate models trained at each level taking the prediction of previous level as an additional input, thereby increasing latency in achieving training accuracy and/or requiring an explicit maintenance module to orchestrate inference and retrain multiple models (corresponding to the number of levels in the hierarchy). The disclosed method and system allows the predefined knowledge about hierarchy drive the learning process of a single model, which predicts all levels of the hierarchy. The disclosed multi-layer network model arrives at a consensus based on prediction at each level, thereby increasing the accuracy of prediction and reducing the training time.
1. A processor-implemented method for product category classification by a multi-layered network model comprising:
defining a domain embedding associated with a plurality of products using a word embedding layer of the multi-layer network, the domain embedding comprising a hierarchy defined based on a plurality of product descriptions associated with the plurality of products;
expressing, by a rule-embedding layer of the multi-layer network, the hierarchy in a bitmap structure as a set of functions associated with a plurality of levels of the hierarchy, the bitmap structure of the hierarchy comprising bitmap values for each element of the plurality of levels of the hierarchy, wherein expressing the hierarchy in the bitmap structure comprises:
defining a parent function fora parent level of the hierarchy by assigning a unique multi-bit value to each element belonging to the parent level, and
defining a child function for subsequent child levels of the parent level based on the parent function of a root level of the hierarchy; and
predicting, for the product category classification, a child level of the hierarchy from amongst the plurality of levels based on a dot product of a bitmap value of the parent level and a filter of the child level.
2. The method of claim 1 , wherein the bitmap values for each element in each of the plurality of levels is precomputed and stored as the filter in matrix format.
3. The method of claim 2 , wherein the child function is defined as:
f (Bitmap(level))=Bitmap(level+1).
4. The method of claim 1 , wherein the rule embedding layer comprises a plurality of dense layers, each of the plurality of dense layers comprising filters for the plurality of hierarchy levels, wherein each filter comprises a custom bitmap layer, the custom bit map layer capable of estimating a probability of occurrence of a set of elements in each level of the plurality of levels of the hierarchy.
5. The method of claim 1 , further comprising determining a confidence score associated with the prediction of the child level using a Softmax function.
6. The method of claim 1 , further comprising connecting the parent level with the child level through a lambda layer, wherein output of the parent level is dependent on an accuracy of the child level.
7. A system ( 500 ) for product category classification by a multi-layered network model, comprising:
one or more memories ( 515 ); and
one or more hardware processors ( 502 ), the one or more memories ( 515 ) coupled to the one or more hardware processors ( 502 ), wherein the one or more hardware processors ( 502 ) are configured to execute programmed instructions stored in the one or more memories ( 515 ), to:
define a domain embedding associated with a plurality of products using a word embedding layer of the multi-layer network, the domain embedding comprising a hierarchy defined based on a plurality of product descriptions associated with the plurality of products;
express, by a rule-embedding layer of the multi-layer network, the hierarchy in a bitmap structure as a set of functions associated with a plurality of levels of the hierarchy, the bitmap structure of the hierarchy comprising bitmap values for each element of the plurality of levels of the hierarchy, wherein expressing the hierarchy in the bitmap structure comprises:
define a parent function fora parent level of the hierarchy by assigning a unique multi-bit value to each element belonging to the parent level, and
define a child function for subsequent child levels of the parent level based on the parent function of a root level of the hierarchy; and
predict, for the product category classification, a child level of the hierarchy from amongst the plurality of levels based on a dot product of a bitmap value of the parent level and a filter of the child level.
8. The system of claim 7 , wherein the bitmap values for each element in each of the plurality of levels is precomputed and stored as the filter in matrix formats.
9. The system of claim 8 , wherein the child function is defined as:
f (Bitmap(level))=Bitmap(level+1).
10. The system of claim 7 , wherein the rule embedding layer comprises a plurality of dense layers, each of the plurality of dense layers comprising filters for the plurality of hierarchy levels, wherein each filter comprises a custom bitmap layer, the custom bit map layer capable of estimating a probability of occurrence of a set of elements in each level of the plurality of levels of the hierarchy.
11. The system of claim 7 , further comprising determining a confidence score associated with the prediction of the child level using a Softmax function.
12. The system of claim 7 , further comprising connecting the parent level with the child level through a lambda layer, wherein output of parent level is dependent of an accuracy of child level.
13. One or more non-transitory machine readable information storage mediums comprising one or more instructions which when executed by one or more hardware processors cause:
defining a domain embedding associated with a plurality of products using a word embedding layer of a multi-layer network, the domain embedding comprising a hierarchy defined based on a plurality of product descriptions associated with the plurality of products;
expressing, by a rule-embedding layer of the multi-layer network, the hierarchy in a bitmap structure as a set of functions associated with a plurality of levels of the hierarchy, the bitmap structure of the hierarchy comprising bitmap values for each element of the plurality of levels of the hierarchy, wherein expressing the hierarchy in the bitmap structure comprises:
defining a parent function fora parent level of the hierarchy by assigning a unique multi-bit value to each element belonging to the parent level, and
defining a child function for subsequent child levels of the parent level based on the parent function of a root level of the hierarchy; and
predicting, for the product category classification, a child level of the hierarchy from amongst the plurality of levels based on a dot product of a bitmap value of the parent level and a filter of the child level.