Building dialogue structure by using communicative discourse trees

Patent metadata

Field	Value
Publication number	US-11977568-B2
Application number	US-202218053909-A
Country	US
Kind code	B2
Filing date	Nov 9, 2022
Priority date	Jan 30, 2018
Publication date	May 7, 2024
Grant date	May 7, 2024

Abstract

Systems, devices, and methods of the present invention detect rhetoric agreement between texts. In an example, a rhetoric agreement application accesses a multi-part initial query and generates a question communicative discourse tree that represents rhetorical relationships between fragments of the query. The application identifies a sub-communication discourse tree from the question communicative discourse tree. The application generates a candidate answer communicative discourse tree for each candidate answer of a set of candidate answers. The application computes a level of complementarity between the sub-discourse tree and each candidate answer discourse tree by applying a classification model to the sub-communication discourse tree and candidate answer communicative discourse trees. The application selects an answer from the candidate answers based on the computed complementarity, thereby building a dialogue structure of an interactive session.

First claim

What is claimed is:

1. A computer-implemented method for training a classification model to predict a complementarity of a pair of two sentences, the method comprising:
accessing a positive dataset and a negative dataset, each dataset comprising training pairs, wherein each training pair comprises (i) a question communicative discourse tree that represents a question and (ii) an answer communicative discourse tree that represents an answer, wherein the positive dataset comprises first training pairs that are above a threshold expected level of complementarity and the negative dataset comprises second training pairs that are below the threshold expected level of complementarity; and
training the classification model by iteratively:
providing one of the training pairs to the classification model,
receiving, from the classification model, a determined level of complementarity;
calculating a loss function by calculating a difference between the determined level of complementarity and the threshold expected level of complementarity; and

adjusting internal parameters of the classification model to minimize the loss function.

2. The method of claim 1 , further comprising:
generating a plurality of complement pair discourse trees; and
assigning each complement pair discourse tree of the plurality of complement pair discourse trees to either the positive dataset or the negative dataset.

3. The method of claim 1 , wherein the negative dataset comprises a question answer pair that comprises an additional question and an additional answer that is relevant but is rhetorically incorrect when compared to the additional question.

4. The method of claim 1 , further comprising:
accessing a question sentence comprising a plurality of fragments, wherein at least one fragment comprises a verb and a plurality of words, and wherein each fragment is an elementary discourse unit;
generating an additional question communicative discourse tree that represents rhetorical relationships between the plurality of fragments and comprises a root node;
identifying a question sub-communication discourse tree from the additional question communicative discourse tree, wherein the question sub-communication discourse tree comprises at least one of the fragments and represents a sub-question;
accessing a plurality of candidate answers, each candidate answer comprising a corresponding plurality of fragments;
generating, for each candidate answer, a candidate answer communicative discourse tree that represents rhetorical relationships between the corresponding plurality of fragments in a respective candidate answer of the plurality of candidate answers, the candidate answer communicative discourse tree comprising a corresponding root node;
for each candidate answer, computing a level of complementarity between the question sub-communication discourse tree and the respective candidate answer discourse tree by applying the classification model to the question sub-communication discourse tree and to the respective candidate answer discourse tree; and
selecting a particular answer from the candidate answers based on the level of complementarity; and
providing, to a user device, the particular answer selected from the candidate answers.

5. The method of claim 4 , wherein identifying the question sub-communicative discourse tree further comprises identifying a particular rhetorical relation that is (i) not joint and (ii) not elaboration.

6. The method of claim 4 , further comprising:
for each candidate answer, computing, using the classification model, a level of rhetorical agreement between the question sub-communicative discourse tree and the respective candidate answer communicative discourse tree; and
selecting the particular answer from the candidate answers further based on the level of rhetorical agreement.

7. The method of claim 1 , wherein accessing a plurality of candidate answers comprises searching for keyword matches derived from elementary discourse units of the question communicative discourse tree against (i) a first database of a discourse corpus, (ii) a second database of a keyword corpus, or (iii) past utterances received.

8. The method of claim 1 , wherein accessing a plurality of candidate answers comprises searching for keyword matches derived from elementary discourse units of the question communicative discourse tree against (i) a first database of a discourse corpus, (ii) a second database of a keyword corpus, or (iii) past utterances received.

9. A system comprising:
a non-transitory computer-readable medium storing computer-executable program instructions; and
a processing device communicatively coupled to the non-transitory computer-readable medium for executing the computer-executable program instructions, wherein executing the computer-executable program instructions causes the processing device to perform operations comprising:
accessing a positive dataset and a negative dataset, each dataset comprising training pairs, wherein each training pair comprises (i) a question communicative discourse tree that represents a question and (ii) an answer communicative discourse tree that represents an answer, wherein the positive dataset comprises first training pairs that are above a threshold expected level of complementarity and the negative dataset comprises second training pairs that are below the threshold expected level of complementarity; and
training a classification model by iteratively:
providing one of the training pairs to the classification model,
receiving, from the classification model, a determined level of complementarity;
calculating a loss function by calculating a difference between the determined level of complementarity and the threshold expected level of complementarity; and

adjusting internal parameters of the classification model to minimize the loss function.

10. The system of claim 9 , wherein executing the computer-executable program instructions further causes the processing device to perform operations comprising:
generating a plurality of complement pair discourse trees; and
assigning each complement pair discourse tree of the plurality of complement pair discourse trees to either the positive dataset or the negative dataset.

11. The system of claim 9 , wherein the negative dataset comprises a question answer pair that comprises an additional question and an additional answer that is relevant but is rhetorically incorrect when compared to the additional question.

12. The system of claim 9 , wherein executing the computer-executable program instructions further causes the processing device to perform operations comprising:
accessing a question sentence comprising a plurality of fragments, wherein at least one fragment comprises a verb and a plurality of words, and wherein each fragment is an elementary discourse unit;
generating an additional question communicative discourse tree that represents rhetorical relationships between the plurality of fragments and comprises a root node;
identifying a question sub-communication discourse tree from the additional question communicative discourse tree, wherein the question sub-communication discourse tree comprises at least one of the fragments and represents a sub-question;
accessing a plurality of candidate answers, each candidate answer comprising a corresponding plurality of fragments;
generating, for each candidate answer, a candidate answer communicative discourse tree that represents rhetorical relationships between the corresponding plurality of fragments in a respective candidate answer of the plurality of candidate answers, the candidate answer communicative discourse tree comprisin

Assignees

• Oracle Int Corp

Inventors

• Galitsky Boris

Classifications

• G06F16/3329 Primary

  Natural language query formulation · CPC title

• G06F16/322

  Trees · CPC title

• G06F16/3349

  Reuse of stored results of previous queries · CPC title

• G06F18/24147

  Distances to closest patterns, e.g. nearest neighbour classification · CPC title

• G06N20/10

  using kernel methods, e.g. support vector machines [SVM] · CPC title