Understanding Inter-Session Intentions via Complex Logical Reasoning

TL;DR

The task of logical session complex query answering (LS-CQA), where sessions are treated as hyperedges of items, is presented and a new model, the Logical Session Graph Transformer (LSGT), which captures interactions among items across different sessions and their logical connections using a transformer structure is introduced.

Abstract

Understanding user intentions is essential for improving product recommendations, navigation suggestions, and query reformulations. However, user intentions can be intricate, involving multiple sessions and attribute requirements connected by logical operators such as And, Or, and Not. For instance, a user may search for Nike or Adidas running shoes across various sessions, with a preference for purple. In another example, a user may have purchased a mattress in a previous session and is now looking for a matching bed frame without intending to buy another mattress. Existing research on session understanding has not adequately addressed making product or attribute recommendations for such complex intentions. In this paper, we present the task of logical session complex query answering (LS-CQA), where sessions are treated as hyperedges of items, and we frame the problem of complex intention understanding as an LS-CQA task on an aggregated hypergraph of sessions, items, and attributes. This is a unique complex query answering task with sessions as ordered hyperedges. We also introduce a new model, the Logical Session Graph Transformer (LSGT), which captures interactions among items across different sessions and their logical connections using a transformer structure. We analyze the expressiveness of LSGT and prove the permutation invariance of the inputs for the logical operators. By evaluating LSGT on three datasets, we demonstrate that it achieves state-of-the-art results.

Figures (5)

• Figure 1: Example complex queries involving varied numbers of sessions, products, and product attributes.
• Figure 2: This figure shows the connections and differences between general hypergraphs, hyper-relational knowledge graphs, and the hyper-session graph in our problem.
• Figure 3: This figure shows the illustration of different query embedding methods. (A) The logical session complex query is expressed in the first-order logic form. (B) The interpretations on the logical session complex query. (C) The computational graph of the complex query proposed by hamilton2018embedding; (D) The linearization of the computational graph to token proposed by bai-etal-2022-query2particles.
• Figure 4: This figure shows the method of LSGT. (A) The computational graph indicates finding the brand of the products that are designed by both session $S_1$ and $S_2$. (B) The node identifiers and type identifiers for the tokens and each of the identifiers is associated with its corresponding embedding vector. (C) The transformer encoder is used for encoding the tokens.
• Figure 5: The query structures are used for training and evaluation. For brevity, the $p$, $i$, $n$, and $u$ represent the projection, intersection, negation, and union operations. The query types are trained and evaluated under supervised settings.

Theorems & Definitions (6)

• Theorem 1
• Theorem 2
• Theorem 3
• Lemma 1
• Lemma 2
• Lemma 3