@prefix rdf:   <http://www.w3.org/1999/02/22-rdf-syntax-ns#> .
@prefix sl:    <http://www.semanlink.net/2001/00/semanlink-schema#> .
@prefix skos:  <http://www.w3.org/2004/02/skos/core#> .
@prefix rdfs:  <http://www.w3.org/2000/01/rdf-schema#> .
@prefix tag:   <http://www.semanlink.net/tag/> .
@prefix foaf:  <http://xmlns.com/foaf/0.1/> .
@prefix dc:    <http://purl.org/dc/elements/1.1/> .

tag:nlp_microsoft  a    sl:Tag ;
        skos:prefLabel  "NLP@Microsoft" .

tag:taxonomies  a       sl:Tag ;
        skos:prefLabel  "Taxonomies" .

tag:entity_alignment  a  sl:Tag ;
        skos:prefLabel  "Entity alignment" .

tag:www_conference  a   sl:Tag ;
        skos:prefLabel  "TheWebConf" .

tag:knowledge_graph_completion
        a               sl:Tag ;
        skos:prefLabel  "Knowledge Graph Completion" .

tag:taxonomy_expansion_task
        a               sl:Tag ;
        skos:prefLabel  "Taxonomy expansion task" .

tag:graph_neural_networks
        a               sl:Tag ;
        skos:prefLabel  "Graph neural networks" .

tag:microsoft_research
        a               sl:Tag ;
        skos:prefLabel  "Microsoft Research" .

<http://www.semanlink.net/doc/2020/03/combining_knowledge_graphs_qui>
        dc:title         "Combining knowledge graphs, quickly and accurately (2020)" ;
        sl:comment       "Entity matching at Amazon: a new [#entity alignment](/tag/entity_alignment) technique that factors in information about the graph in the vicinity of the entity name.\r\n\r\n[#Graph neural network](/tag/graph_neural_networks) that specifically addresses the problem of **merging multi-type knowledge graphs**. " ;
        sl:creationDate  "2020-03-19" ;
        sl:tag           tag:thewebconf_2020 , tag:graph_neural_networks , tag:entity_alignment , tag:combining_knowledge_graphs , tag:attention_knowledge_graphs , tag:ai_amazon .

tag:thewebconf_2020  a    sl:Tag ;
        rdfs:isDefinedBy  tag:thewebconf_2020.n3 ;
        skos:broader      tag:www_conference ;
        skos:prefLabel    "TheWebConf 2020" ;
        foaf:page         tag:thewebconf_2020.html .

tag:ai_amazon  a        sl:Tag ;
        skos:prefLabel  "AI@Amazon" .

tag:discute_avec_raphael
        a               sl:Tag ;
        skos:prefLabel  "Discuté avec Raphaël" .

tag:attention_knowledge_graphs
        a               sl:Tag ;
        skos:prefLabel  "Attention + Knowledge Graphs" .

tag:text_aware_kg_embedding
        a               sl:Tag ;
        skos:prefLabel  "Text-Aware KG embedding" .

tag:arxiv_doc  a        sl:Tag ;
        skos:prefLabel  "Arxiv Doc" .

<http://www.semanlink.net/doc/2020/04/2001_09522_taxoexpan_self_su>
        dc:title         "[2001.09522] TaxoExpan: Self-supervised Taxonomy Expansion with Position-Enhanced Graph Neural Network" ;
        sl:comment       "how to add a set of new concepts to an existing taxonomy. \r\n\r\n[Tweet](https://twitter.com/mickeyjs6/status/1253772146142216194?s=20) [GitHub](https://github.com/mickeystroller/TaxoExpan)\r\n\r\n> we study the taxonomy expansion task: given an\r\nexisting taxonomy and a set of new emerging concepts, we aim\r\nto automatically expand the taxonomy to incorporate these new\r\nconcepts (without changing the existing relations in the given taxonomy).\r\n\r\n> To the best of our knowledge, this is the first study on **how to\r\nexpand an existing directed acyclic graph (as we model a taxonomy\r\nas a DAG) using self-supervised learning**.\r\n\r\nSelf-supervised framework, the existing taxonomy being used as training data: it learns a model to predict whether a query concept is the direct hyponym of an anchor concept. \r\n\r\n> 2 techniques:\r\n>\r\n> 1. a **position-enhanced graph neural network that encodes the local structure of an anchor concept** in the existing taxonomy,\r\n> 2. a noise-robust training objective that enables the learned model to be insensitive to the label noise in the self-supervision data. \r\n\r\nRegarding 1: uses [GNN](/tag/graph_neural_networks.html) to model the \"ego network\" of concepts (potential “siblings”\r\nand “grand parents” of the query concept).\r\n\r\n> Regular\r\nGNNs fail to distinguish nodes with different relative positions to\r\nthe query (i.e., some nodes are grand parents of the query while\r\nthe others are siblings of the query). To address this limitation, we\r\npresent a simple but effective enhancement to inject such position\r\ninformation into GNNs using position embedding. We show that\r\nsuch embedding can be easily integrated with existing GNN architectures\r\n(e.g., [GCN](/tag/graph_convolutional_networks) and GAT) and significantly boosts the\r\nprediction performance\r\n\r\nRegarding point 2: uses InfoNCE loss, cf. [Contrastive Predictive Coding](/doc/?uri=https%3A%2F%2Farxiv.org%2Fabs%2F1807.03748)\r\n\r\n> Instead of predicting\r\nwhether each individual ⟨query concept, anchor concept⟩ pair\r\nis positive or not, we first group all pairs sharing the same query\r\nconcept into a single training instance and learn a model to select\r\nthe positive pair among other negative ones from the group. \r\n\r\n(Hum, ça me rappelle quelque chose)\r\n\r\n> assume each concept (in existing taxonomy + set of new concepts) has an initial embedding\r\nvector learned from some text associated with this concept.\r\n\r\nTo keep things tractable, only attempts to find a single parent node of each new concept." ;
        sl:creationDate  "2020-04-25" ;
        sl:tag           tag:taxonomies , tag:text_aware_kg_embedding , tag:microsoft_research , tag:discute_avec_raphael , tag:taxonomy_expansion_task , tag:nlp_microsoft , tag:graph_neural_networks , tag:arxiv_doc , tag:knowledge_graph_completion , tag:thewebconf_2020 .

tag:combining_knowledge_graphs
        a               sl:Tag ;
        skos:prefLabel  "Combining knowledge graphs" .