Tailor knowledge graph for query understanding: linking intent topics by propagation

Abstract

Knowledge graphs are recently used for enriching query representations in an entity-aware way for the rich facts organized around entities in it.However, few of the methods pay attention to non-entity words and clicked websites in queries, which also help conveying user intent. In this paper, we tackle the problem of intent understanding with innovatively representing entity words, refiners and clicked urls as intent topics in a unified knowledge graph based framework, in a way to exploit and expand knowledge graph which we call 'tailor'. We collaboratively exploit global knowledge in knowledge graphs and local contexts in query log to initialize intent representation, then propagate the enriched features in a graph consisting of intent topics using an unsupervised algorithm. The experiments prove intent topics with knowledge graph enriched features significantly enhance intent understanding.

VISTA: Visual-Textual Knowledge Graph Representation Learning

“…While some knowledge graph embedding methods utilize images of entities (Xie et al, 2017;Liu et al, 2019), some recently proposed multimodal methods consider both images and text descriptions of entities (Pezeshkpour et al, 2018;Wang et al, 2019). For example, MoSE (Zhao et al, 2022) and IMF (Li et al, 2023) learn modalityspecific representations and make predictions using the representations from different modalities. Also, OTKGE (Cao et al, 2022) proposes an optimal transport to align multi-modal embeddings, while MKGformer (Chen et al, 2022) conducts multi-level fusion using a hybrid transformer.…”

“…We use four datasets shown in Table 2; VTKG-I and VTKG-C are real-world VTKG datasets introduced in Section 3.2. While WN18 (Bordes et al, 2013) and FB15K237 (Toutanova and Chen, 2015) are benchmark datasets used in other multimodal knowledge graph completion research (Zhao et al, 2022;Chen et al, 2022), WN18 has a test leakage issue, and WN18RR (Dettmers et al, 2018) has been proposed to resolve the issue. In our experiments, we use WN18RR++ which is the fixed version of WN18RR as described in Section 3.2.…”

“…We use ten different baseline methods: ANAL-OGY , ComplEx-N3 (Lacroix et al, 2018), RotatE (Sun et al, 2019), PairRE (Chao et al, 2021), RSME , TransAE (Wang et al, 2019), MKGformer (Chen et al, 2022), OTKGE (Cao et al, 2022), MoSE (Zhao et al, 2022), and IMF (Li et al, 2023) We divide the training, valid, and test sets with a ratio of 8:1:1 for VTKGs; we use the provided split for the existing benchmark datasets. When training a model using triplets' images, we only use the images associated with the triplets in a training set.…”

“…While most existing knowledge graph embedding methods focus solely on the structure of a knowledge graph to learn the representations (Lacroix et al, 2018;Sun et al, 2019), using additional images or text descriptions can result in much better representations. A few * Corresponding author recently proposed multimodal knowledge graph completion methods (Zhao et al, 2022;Li et al, 2023) consider the case where entities can have their images and text descriptions; Figure 1 (a) is a form of knowledge graph existing multimodal knowledge graph completion methods consider. We focus on the fact that some triplets can be more intuitively expressed by their images.…”

Graph Enhanced BERT for Query Understanding

“…search log) or external sources (e.g. Knowledge Graph) has been leveraged to address this challenge [38]. Second, the majority of existing solutions [2,5] often train models from scratch in a supervised way, which require a large number of taskspecific labeled data.…”