Identifying and ranking potentially privileged documents using a machine learning topic model
A method for identifying and ranking potentially privileged documents using a machine learning topic model may include receiving a set of documents. The method may also include, for each of two or more documents in the set of documents, extracting a set of spans from the document, generating, using a machine learning topic model, a set of topics and a subset of legal topics for the set of spans, generating a vector of probabilities for each span with a probability being assigned to each topic in the set of topics for the span, assigning a score to one or more spans in the set of spans by summing the probabilities in the vector that are assigned to a topic in the subset of legal topics, and assigning a score to the document. The method may further include ranking the two or more documents by their assigned scores.
1 . A computer-implemented method for identifying and ranking potentially privileged documents using a machine learning topic model, at least a portion of the method being performed by a computer system comprising one or more processors, the method comprising:
receiving, by the computer system, a document;
generating a set of topics for the document by the computer system using a machine learning topic model selected from a group of machine learning topic models consisting of: Latent Dirichlet Allocation, Latent Semantic Analysis, Correlated Topic Modeling, Probabilistic Latent Semantic Analysis, non-negative matrix factorization, Parallel Latent Dirichlet Allocation, Pachinko Allocation Model, lda2vec, Gibbs Sampling Dirichlet Multinomial Mixture, Dirichlet Multinomial Mixture, and derivations thereof;
generating, by the computer system, a vector of probabilities, each of the probabilities corresponding to a topic in the set of topics; and
assigning, by the computer system, a score to the document based on the probabilities in the vector that are assigned to a subset of the topics, the score indicating a potential that the document is privileged.
2 . The method of claim 1 , further comprising:
labelling, by the computer system, the document as potentially privileged if the document has a score above a certain threshold.
3 . The method of claim 1 , further comprising:
generating, by the computer system, a master set of topics for the document, the set of topics being a subset of the master set of topics.
4 . The method of claim 1 , further comprising:
altering, by the computer system, the score assigned to a first span if the first span contains a mention of a legal entity.
5 . The method of claim 1 , further comprising:
altering, by the computer system, the score assigned to the document if the document contains a mention of a legal entity.
6 . The method of claim 1 , further comprising:
extracting, by the computer system, a set of spans from the document,
wherein generating a set of topics for the document includes generating a set of topics for each span of the set of spans,
wherein generating the vector of probabilities includes generating a vector of probabilities for each span in the set of spans, and
assigning, by the computer system, a score to one or more spans in the set of spans by mathematically combining the probabilities in the vector that are assigned to a subset of the topics,
wherein assigning the score to the document is based on the scores assigned to the one or more spans in the set of spans.
7 . The method of claim 6 , wherein the assigning, by the computer system, of a score to the one or more spans in the set of spans is not performed for a first span of the one or more spans in the set of spans if at least one of: a language of the first span is unknown or cannot be detected, a number of tokens in the first span is less than a threshold, and the number of tokens in the first span is less than the threshold after names are removed from the first span.
8 . The method of claim 6 , further comprising:
altering, by the computer system, the score assigned to a first span if a sender or a recipient of a document containing the first span is a legal entity.
9 . The method of claim 1 , further comprising:
receiving, by the computer system, a set of documents that includes the document;
for each document in the set of documents:
generating a set of topics for the document by the computer system using a machine learning topic model selected from a group of machine learning topic models consisting of: Latent Dirichlet Allocation, Latent Semantic Analysis, Correlated Topic Modeling, Probabilistic Latent Semantic Analysis, non-negative matrix factorization, Parallel Latent Dirichlet Allocation, Pachinko Allocation Model, lda2vec, Gibbs Sampling Dirichlet Multinomial Mixture, Dirichlet Multinomial Mixture, and derivations thereof;
obtaining, by the computer system, a vector of probabilities, each of the probabilities corresponding to a topic in the set of topics; and
assigning, by the computer system, a score to the document by summing the probabilities in the vector that are assigned to a subset of the topics, the score indicating a potential that the document is privileged;
creating, by the computer system, a hypergraph comprising a set of entities based on communications between the set of entities in the set of documents; and
weighting, by the computer system, an edge between one or more entities in the set of entities if the communications between the one or more entities is potentially privileged; and
altering, by the computer system, the score assigned to a second document in the set of documents if the second document is a communication between entities with a weighted edge in the hypergraph.
10 . One or more non-transitory computer-readable media comprising one or more computer-readable instructions that, when executed by one or more processors of a computer system, cause the computer system to perform the method of claim 1 .
11 . A computer-implemented method for identifying and ranking potentially privileged documents using a machine learning topic model, at least a portion of the method being performed by a computer system comprising one or more processors, the method comprising:
receiving, by the computer system, a document;
generating, by the computer system, a first vector that includes words of the document;
applying, by the computer system, a machine learning topic model to the first vector to generate a set of topics for the document;
generating, by the computer system, a second vector of probabilities, each of the probabilities corresponding to a topic in the set of topics; and
assigning, by the computer system, a score to the document based on the probabilities in the second vector that are assigned to a subset of the topics, the score indicating a potential that the document is privileged.
12 . The method of claim 11 , further comprising:
labelling, by the computer system, the document as potentially privileged if the document has a score above a certain threshold.
13 . The method of claim 11 , further comprising:
generating, by the computer system, a master set of topics for the document, the set of topics being a subset of the master set of topics.
14 . The method of claim 11 , further comprising:
altering, by the computer system, the score assigned to a first span if the first span contains a mention of a legal entity.
15 . The method of claim 11 , further comprising:
altering, by the computer system, the score assigned to the document if the document contains a mention of a legal entity.
16 . The method of claim 11 , further comprising:
extracting, by the computer system, a set of spans from the document,
wherein generating a set of topics for the document includes generating a set of topics for each span of the set of spans,
wherein generating the second vector of probabilities includes generating a second vector of probabilities for each span in the set of spans; and
assigning, by the computer system, a score to one or more spans in the set of spans by mathematically combining the probabilities in the second vector that are assigned to a subset of the topics,
wherein assigning the score to the document is based on the scores assigned to the one or more spans in the set of spans.
17 . The method of claim 16 , wherein the assigning, by the computer system, of a score to the one or more spans in the set of spans is not performed for a first span of the one or more spans in the set of spans if at least one of: a language of the first span is unknown or cannot be detected, a number of tokens in the first span is less than a threshold, and the number of tokens in the first span is less than the threshold after names are removed from the first span.
18 . The method of claim 11 , further comprising:
receiving, by the computer system, a set of documents that includes the document;
for each document in the set of documents:
generating a set of topics for the document by the computer system using a machine learning topic model selected from a group of machine learning topic models consisting of: Latent Dirichlet Allocation, Latent Semantic Analysis, Correlated Topic Modeling, Probabilistic Latent Semantic Analysis, non-negative matrix factorization, Parallel Latent Dirichlet Allocation, Pachinko Allocation Model, lda2vec, Gibbs Sampling Dirichlet Multinomial Mixture, Dirichlet Multinomial Mixture, and derivations thereof;
obtaining, by the computer system, a second vector of probabilities, each of the probabilities corresponding to a topic in the set of topics; and
assigning, by the computer system, a score to the document by summing the probabilities in the second vector that are assigned to a subset of the topics, the score indicating a potential that the document is privileged;
creating, by the computer system, a hypergraph comprising a set of entities based on communications between the set of entities in the set of documents; and
weighting, by the computer system, an edge between one or more entities in the set of entities if the communications between the one or more entities is potentially privileged; and
altering, by the computer system, the score assigned to a second document in the set of documents if the second document is a communication between entities with a weighted edge in the hypergraph.
19 . One or more non-transitory computer-readable media comprising one or more computer-readable instructions that, when executed by one or more processors of a computer system, cause the computer system to perform the method of claim 11 .
20 . A system comprising:
one or more processors; and
one or more non-transitory computer-readable media comprising one or more computer-readable instructions that, when executed by the one or more processors, cause the system to perform operations, the operations comprising:
receiving a document;
generating a set of topics for the document using a machine learning topic model selected from a group of machine learning topic models consisting of: Latent Dirichlet Allocation, Latent Semantic Analysis, Correlated Topic Modeling, Probabilistic Latent Semantic Analysis, non-negative matrix factorization, Parallel Latent Dirichlet Allocation, Pachinko Allocation Model, lda2vec, Gibbs Sampling Dirichlet Multinomial Mixture, Dirichlet Multinomial Mixture, and derivations thereof;
generating a vector of probabilities, each of the probabilities corresponding to a topic in the set of topics; and
assigning a score to the document based on the probabilities in the vector that are assigned to a subset of the topics, the score indicating a potential that the document is privileged.