Alibaba Innovative Research (AIR) > Natural Language Processing
Cognitive Knowledge Graph in E-Commerce

Theme

Natural Language Processing

Topic

Cognitive Knowledge Graph in E-Commerce

Background

One of the ultimate goals of e-commerce platforms is to satisfy various shopping needs for their customers. Much efforts are devoted to creating taxonomies or ontologies in e-commerce towards this goal. However, user needs in e-commerce are never well defined, and none of the existing ontologies has the enough depth and breadth for universal user needs understanding. The semantic gap in-between prevents shopping experience from being more intelligent. 

Therefore, we propose to construct a large-scale e-commerce Cognitive Concept net named “AliCoCo”, which is practiced in Alibaba, the largest Chinese e-commerce platform in the world. Different from traditional e-commerce KGs, which mainly describe the attributes and relations of items and can be regarded as "Product Graphs", we formally define user needs in e-commerce, and conceptualize user needs as explicit concept nodes in the graph. For example, "Outdoor barbecue" and "Keep Warm for Kids" are such concept in AliCoCo. Besides concepts, nodes also include open-domain entities such as famous persons or places, e-commerce entities such as brands and items, etc. Relations include "isA", "hypernym/hyponym", "related_to" and other open-domain relations. Based on AliCoCo, various downstream applications such as e-commerce search and recommendation can be benefited.  

Great efforts are already devoted to the construction of the initial version of AliCoCo, and we believe there are plenty of research topics worth digging for further upgrading AliCoCo.

Target 

  • A well-defined ontology in e-commerce, which is able to describe the complex relations of users, items, shopping needs and commonsense knowledge in the scenario of multi-lingual e-commerce.
  • A system of automatic extraction of entities, concepts and relations from both open-domain and vertical domains related to e-commerce.
  • A thorough understanding of downstream applications (e.g. e-commerce search, recommendation and advertising) of our knowledge graph.

Related Research Topics 

  • Self-learned Information Extraction via Online Feedbacks. Traditional information extraction (IE) refers to the extraction of relation tuples from plain text via supervised learning from massive labeled data.  In e-commerce platform, online feedbacks from real-world used are abundant and never utilized to supervise offline tasks. Thus, we are interested in using online feedbacks to help offline extraction models.
  • Commonsense Knowledge Mining in E-commerce. E-commerce is a vertical domain which may require a lot of commonsense knowledge, such as complementary relations between two categories, mapping relations between informal queries to standard product attributes. 
  • Knowledge Fusion and Refinement for E-commerce Knowledge Graph.As introduced above, nodes in AliCoCo belong to different types, such as categories, brands, basic concepts, combinational concepts, etc.  Besides e-commerce knowledge, it is necessary to introduce more open-domain or commonsense knowledge from WordNet, Freebase, ConceptNet, DBpedi, etc. After knowledge fusion, we need to refine nodes and edges by disambiguation and alignment using domain context.
  • Knowledge aware Text/Query Understanding. AliCoCo contains rich and valuable semantic information in e-commerce scenario. Linking entities in user search queries or other e-commerce text to our concept net can further benefit user needs understanding and improve downstream applications.
  • Knowledge Representation & Reasoning. Knowledge representation and reasoning aims to represent entities and relations of a knowledge graph in low-dimensional semantic space. AliCoCo contains rich and valuable semantic information. Using such technique to embed nodes and edges and further reason with application systems such as e-commerce search engine and recommender system.

 

Suggested Collaboration Method

AIR (Alibaba Innovative Research), one-year collaboration project. 

 

 

Scan QR code
关注Ali TechnologyWechat Account