System and method for handling popularity bias in item recommendations

A system (600) comprising:

• one or more memories (604);
• and
• one or more first hardware processors (602), the one or more first memories (604) coupled to the one or more first hardware processors (602), wherein the one or more first hardware processors (602) are configured to execute programmed instructions stored in the one or more first memories (604), to:
• initialize an item embedding look-up matrix corresponding to a plurality of items in a sequence of item-clicks associated with a training data;
• apply L2 norm to the item embedding look-up matrix to learn normalized item embeddings corresponding to the item embedding look-up matrix;
• model, using a neural network (NN), session embeddings corresponding to the sequences of item-clicks;
• apply L2 norm to the session embeddings to obtain normalized session embeddings;
• obtain relevance score corresponding to each of the plurality of items based on similarity between the normalized item embeddings and the normalized session embeddings;
• estimate a multi-dimensional probability vector corresponding to the relevance scores for the plurality of items to be clicked in the sequence;
• and
• recommend a list of the plurality of items ordered based on the multi-dimensional probability vector.

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:

• initializing, via one or more hardware processors, an item embedding look-up matrix corresponding to a plurality of items in a sequence of item-clicks associated with a training data;
• applying, via the one or more hardware processors, L2 norm to the item embedding look-up matrix to learn normalized item embeddings corresponding to the item embedding look-up matrix;
• modelling, using a neural network (NN), session embeddings corresponding to the sequence of item-clicks, via the one or more hardware processors;
• applying L2 norm to the session embeddings to obtain normalized session embeddings, via the one or more hardware processors;
• obtaining, via the one or more hardware processors, relevance score corresponding to each of the plurality of items based on a similarity between the normalized item embeddings and the normalized session embeddings;
• estimating, via the one or more hardware processors, a multi-dimensional probability vector corresponding to the relevance scores for the plurality of items to be clicked in the sequence;
• and
• recommending, via the one or more hardware processors, a list of the plurality of items ordered based on the multi-dimensional probability vector.