Back to results
US 9,519,871 B1Granted
Contextual text adaptation
AssigneeInternational Business Machines Corporation
InventorsCarlos H. Cardonha, Stefany Mazon, Debora L. Nascimento Ciriaco Pereira, Nicole B. Sultanum
Life of this patent
Estimating the term…
Abstract
Contextual adaptation of documents automatically replaces words for synonyms that appear within context or topic whey they are being used. A machine learned topic modeling, trained by a set of documents representative of a target user is executed to determine topics of an input document, and to determine words in the document to replace based on determining the relevance of the words to the topics in the documents. An output document is generated based on the input document with the replaced words.
Claims18
A system for contextual text adaptation, comprising:
- one or more hardware processors;
- a topic model algorithm executable on one or more of the hardware processors, the topic model algorithm generated by machine learning based on a corpus of documents related to context of a target user and a dictionary, the topic model comprising a first function that predicts probability distribution of a plurality of topics in a given document, and a second function that predicts probability of a given word occurring in a document associated with a given topic, the dictionary comprising a list of words considered for substitution,
- one or more of the hardware processors operable to receive an input document,
- one or more of the hardware processors further operable to determine input document topics associated with the input document and a normalized weight associated with each of the input document topics by executing the first function,
- one or more of the hardware processors further operable to determine an aggregate probability indicating relevance of an input document word to the input document topics based on executing the second function,
- one or more of the hardware processors further operable to determine a synonym of the input document word based on a dictionary of synonyms,
- one or more of the hardware processors further operable to determine an aggregate probability for the synonym based on executing the second function,
- one or more of the hardware processors further operable to compare the aggregate probability for the synonym and the aggregate probability for the input document word, and
- responsive to determining that the aggregate probability for the synonym is greater than the aggregate probability for the input document word, one or more of the hardware processors further operable to replace the input document word with the synonym,
- one or more of the hardware processors further operable to generate an output document comprising content of the input document with replaced word.
7 more elements
A computer-implemented method of contextual text adaptation, the method performed by one or more hardware processors, comprising:
- receiving a corpus of documents in context of a target user;
- receiving a dictionary of words;
- receiving a dictionary of synonyms;
- generating a topic model algorithm based on the corpus of documents and the dictionary of words by machine learning, the topic model algorithm comprising a first function that predicts probability distribution of a plurality of topics in a given document, and a second function that predicts probability of a given word occurring in a document associated with a given topic;
- and
- storing the first function and the second function of the topic model algorithm in a storage device,
- receiving an input document;
- determining input document topics associated with the input document and a normalized weight associated with each of the input document topics by executing the first function;
- determining an aggregate probability indicating relevance of an input document word to the input document topics based on executing the second function;
- determining a synonym of the input document word based on the dictionary of synonyms;
- determining an aggregate probability for the synonym based on executing the second function;
- comparing the aggregate probability for the synonym and the aggregate probability for the input document word;
- and
- responsive to determining that the aggregate probability for the synonym is greater than the aggregate probability for the input document word, replacing the input document word with the synonym;
- and
- generating an output document comprising content of the input document with replaced word.
13 more elements
A computer readable storage medium storing a program of instructions executable by a machine to perform a method of contextual text adaptation, the method comprising:
- identifying a target user;
- receiving a corpus of documents in context of the target user;
- receiving a dictionary of words;
- receiving a dictionary of synonyms;
- generating a topic model algorithm based on the corpus of documents and the dictionary of words by machine learning, the topic model algorithm comprising a first function that predicts probability distribution of a plurality of topics in a given document, and a second function that predicts probability of a given word occurring in a document associated with a given topic;
- and
- storing the first function and the second function of the topic model algorithm in a storage device,
- receiving an input document;
- determining input document topics associated with the input document and a normalized weight associated with each of the input document topics by executing the first function;
- for each of a plurality of input document words in the input document,
- determining a probability that an input document word is associated with an input document topic for each of the input document topics by executing the second function;
- determining an aggregate probability for the input document word as a sum of products of the probability that an input document word is associated with an input document topic and the normalized weight of the input document topic;
- determining a synonym of the input document word based on the dictionary of synonyms;
- determining an aggregate probability for the synonym;
- comparing the aggregate probability for the synonym and the aggregate probability for the input document word;
- responsive to determining that the aggregate probability for the synonym is greater than the aggregate probability for the input document word, replacing the input document word with the synonym;
- and
- generating an output document comprising content of the input document with replaced words.
15 more elements
Dependencies are inferred from the claim text.