Download presentation
Presentation is loading. Please wait.
Published byTimothy Kennedy Modified over 7 years ago
1
Reasoning With Neural Tensor Networks for Knowledge Base Completion
Author: Richard Socher Presented by Batselem 숭실대학교
2
Overview Introduction to Knowledge Base Knowledge Base Reasoning
Knowledge Base Completion Neural Tensor Network Knowledge Base Completion Experiment
3
Knowledge Base A knowledge base is a special kind of database for knowledge management. Designed for storing complex information Captures human knowledge and places it into a computer system Used to solve complex problems that require a high level of human expertise. Knowledge from an expert Non-expert user User Interface query Inference Engine (추론 시스템) Knowledge base advice “Captures human knowledge and places it into a computer system where it is used to solve complex problems requiring a high level of human expertise.“ “Knowledge is Relevant information that one is able to recall from memory. A database is a structured collection of records or data” A collection of information organized in such a way that a computer program can quickly select desired pieces of data. It is possible to make conclusions from the information held in the knowledge base.
4
Knowledge Base Reasoning
Knowledge base consists of a large amount of facts (triples) Represented as triples of a relation between two entities T = (e1, R, e2) Reasoning over a large knowledge base (KB) Entity1 Entity2 Relation Shakespeare Hamlet authorOf Knowledge Base Facts: ex:book1 rdf:type ex:Publication . ex:book2 rdf:type ex:Article . ex:Article rdfs:subClassOf ex:Publication . ex:publishes rdfs:range ex:Publication . ex:MITPress ex:publishes ex:book3 . … rdf:type ex:book1 ex:Publication rdfs:subClassOf ex:book2 rdfs:range rdf:type ex:Article ex:publishes ex:book3 ex:MITPress
5
Knowledge Base Reasoning
Knowledge base consists of a large amount of facts (triples) Represented as triples of a relation between two entities T = (e1, R, e2) Reasoning over a large knowledge base (KB) Entity1 Entity2 Relation Shakespeare Hamlet authorOf Logic (IF-THEN) Rules: IF X is subclass of Y and Z is type of X THEN Z is type of Y rdf:type ex:book1 ex:Publication Infer new relations rdf:type rdfs:subClassOf rdf:type ex:book2 rdfs:range rdf:type ex:Article ex:publishes ex:book3 ex:MITPress
6
Knowledge Base Completion
Knowledge bases are an important resource for question answering and other natural language processing (NLP) tasks Question Answering (Apple Siri, IBM Watson), Information Retrieval Providing structured knowledge to users. Problems with KB: Suffer from incompleteness Lack of ability to reason over their discrete entities and relationships This paper adopts the goal of predicting the likely truth of additional facts based on existing facts in the knowledge base predict KB: Existing facts KB: Additional facts
7
Knowledge Base Completion
Introduce a model that can accurately predict additional true facts using only an existing database. (e1, R, e2) = (Ben Affleck, occupation, FilmMaker) is true or false Score Argo Violet Anne Relation director Neural Tensor Network playedIn childOf Actor Ben Affleck gender occupation male livedIn occupation won FilmMaker (Tom cruise, playedIn, Mission Impossible) (Tom cruise, occupation, Actor) (Matt Damon, playedIn, Martian) (Matt Damon, occupation, Actor) …. Los Angeles Oscar Award
8
Neural Tensor Network Word2vec
Each relation is described by a neural network and pairs of entities are given as input to the model. Each entity has a vector representation, which can be constructed by its word vectors. The model returns a high score if they are in that relationship and a low one otherwise. This allows any fact, whether implicitly or explicitly mentioned in the database to be answered with a certainty score. Large text data Word2vec
9
Neural Tensor Network It represents each entity as the average of its word vectors, . an entity vector as the composition of its word vectors. For example The goal of their approach is to be able to state whether two entities (e1, e2) are in a certain relationship R. For instance, whether the relationship (e1, R, e2) = (Bengal tiger, has part, tail) is true and with what certainty. To this end, they define a set of parameters indexed by R for each relation’s scoring function VectorBank_of_China VectorBank_of_China= 0.5*(VectorBank+ VectorChina) VectorBank VectorChina
10
Neural Tensor Network The Neural Tensor Network (NTN) replaces a standard linear neural network layer with a bilinear tensor layer that directly relates the two entity vectors across multiple dimensions. The model computes a score of how likely it is that two entities are in a certain relationship by the following NTN-based function g(e1, R, e2): f = tanh is a standard nonlinearity applied element-wise WR[1:k] ∈ ℛ d×d×k is a tensor and the bilinear tensor product Standard layer weight VR ∈ ℛ k×2d
11
Neural Tensor Network R
Reference:An MLP-based representation of neural tensor networks for the RDF data models
12
Neural Tensor Network Training objective: T(i)c = (e(i)1, R(i),ec) is a triplet with a random entity corrupted from a correct triplet T (i) = (e(i)1, R(i), e(i)2) Score the correct relation triplet higher than its corrupted one up to a margin of 1. For example, given a correct triplet (Pablo Picaso, nationality, Spain), a potential negative example is (Pablo Picaso, nationality, United States). For each correct triplet we sample C random corrupted triplets.
13
Neural Tensor Network The model is trained by taking derivatives with respect to the five groups of parameters. Cost Function Update W using Backpropagation: 𝐽 𝜕𝐽/𝜕𝑊 𝑙𝑜𝑐𝑎𝑙 𝑚𝑖𝑛𝑖𝑚𝑢𝑚 𝑝𝑜𝑖𝑛𝑡 𝑊=𝑊 − 𝜕𝐽/𝜕𝑊
14
Experiment Datasets Relation Triplets Classification
They randomly switch entities from correct testing triplets resulting in an equal number of positive and negative examples. We predict the relation (e1, R, e2) holds if g(e1, R, e2) ≥ TR
15
Experiment
16
Experiment Given place of birth is Florence and profession is historian, relation model can accurately predict that Francesco Guicciardini’s gender is male and his nationality is Italy. These might be inferred from two pieces of common knowledge: (i) Florence is a city of Italy; (ii) Francesco is a common name among males in Italy.
17
Q&A Thank you
Similar presentations
© 2025 SlidePlayer.com Inc.
All rights reserved.