Note: Descriptions are shown in the official language in which they were submitted.
CA 03176868 2022-09-23
INTENT IDENTIFYING METHOD AND DEVICE FOR APPLICATION TO
INTELLIGENT CUSTOMER SERVICE ROBOT
BACKGROUND OF THE INVENTION
Technical Field
[0001] The present invention relates to the field of artificial intelligence
technology, and more
particularly to an intent identifying method for application to an intelligent
customer
service robot, and a corresponding device.
Description of Related Art
[0002] With the rapid development of businesses, artificial intelligence
technology has been
progressing by leaps and bounds, and the advent of customer service robots
effectively
shares the workload of human customer service, economizes on personnel cost of
enterprises, breaks restrictions in time, manpower, and regionalization,
supplies
uninterrupted consultation services for 24 hours a day and seven days a week,
and
alleviates pain points of human customer service. Customer service robots can
accept
various questions raised by users, and one of the keys bestowing highly
effective
availability to a customer service robot is whether it can judge out the true
intent of a user
according to information provided by the user.
[0003] An extremely rapid development of customer service robots has been seen
over the recent
years notwithstanding, since user-interactive data is involved therein, higher
sensitivity
is present, and there are very few texts with intents in all the dialogue
texts, so the
traditional intent identification work is faced with many such challenges as
the chat texts
are semantically understood not much deeply, and it is impossible to quickly
and
accurately comprehend the user intent in a shorter dialogue text with the
user.
1
Date Regue/Date Received 2022-09-23
CA 03176868 2022-09-23
[0004] Accordingly, it is a problem to be urgently dealt with as how to ensure
that the intelligent
customer service robot quickly and accurately comprehends the user intent, so
as to
quickly and precisely answer questions raised by the user.
SUMMARY OF THE INVENTION
[0005] In view of the above, embodiments of the present invention provide an
intent identifying
method for application to an intelligent customer service robot, and a
corresponding
device, to realize quick and accurate recognition of user intent by the
intelligent customer
service robot, and to supply guarantee for the robot to accurately answer
questions raised
by the user.
[0006] Technical solutions provided by the embodiments of the present
invention are as follows.
[0007] According to the first aspect, there is provided an intent identifying
method for
application to an intelligent customer service robot, and the method comprises
the
following steps:
[0008] SO ¨ obtaining a dialogue text of a user;
[0009] S2 ¨ judging whether the dialogue text contains any intent, if yes,
executing step S4, if
not, terminating the process, if impossible to judge, executing step S3;
[0010] S3 ¨ contextually expanding the dialogue text, and executing step S4
after step S3;
[0011] S4 ¨ identifying a named entity set in the dialogue text, and
determining any intent
knowledge point associated with the named entity set;
[0012] S5 ¨ expressing the dialogue text in distributed word vectors, and
employing plural pre-
trained semantically classifying models for prediction to obtain plural pieces
of semantic
information; and
[0013] S6 ¨ employing an Ensemble framework to merge and tune the intent
knowledge point
and the plural pieces of semantic information, and obtaining a user intent.
2
Date Regue/Date Received 2022-09-23
CA 03176868 2022-09-23
[0014] Moreover, prior to step S2, the method further comprises the following
step:
[0015] Si ¨ performing text rectification on the dialogue text.
[0016] Further, step Si specifically includes:
[0017] term-segmenting the dialogue text, and identifying any erroneous
segmented term in the
dialogue text;
[0018] obtaining a rectifying term to which the erroneous segmented term
corresponds; and
[0019] replacing the erroneous segmented term in the dialogue text with the
rectifying term.
[0020] Further, step S3 specifically includes:
[0021] storing user conversation information with one session as a unit;
[0022] associating with contextual information of the dialogue text, and
judging whether the user
intent is changed, wherein the contextual information includes an intent
identifying result
of context of the dialogue text; and
[0023] employing a near-synonym of the context to expand the dialogue text
when the user intent
is not changed.
[0024] Further, step S4 specifically includes:
[0025] performing a term-segmenting process on the dialogue text according to
a preset
dictionary, and obtaining plural segmented terms;
[0026] matching the plural segmented terms with a preset entity lexicon, and
obtaining the
named entity set; and
[0027] determining an intent knowledge point relevant to the named entity set
from a preset
knowledge base.
[0028] Further, step S5 specifically includes:
[0029] performing a term-segmenting process on the dialogue text, and
obtaining plural
segmented terms;
3
Date Regue/Date Received 2022-09-23
CA 03176868 2022-09-23
[0030] calculating word vectors of the plural segmented terms, and expressing
the word vectors
of the plural segmented terms in distribution; and
[0031] inputting the word vectors of the plural segmented terms expressed in
distribution to the
plural semantically classifying models to output the plural pieces of semantic
information.
[0032] Further, step S6 specifically includes:
[0033] determining a final user intent through the Ensemble framework
according to the intent
knowledge point, the plural pieces of semantic information, and preset weights
to which
the plural pieces of semantic information respectively correspond.
[0034] According to the second aspect, there is provided an intent identifying
device for
application to an intelligent customer service robot, and the device
comprises:
[0035] a text obtaining module, for obtaining a dialogue text of a user;
[0036] an intent judging module, for judging whether the dialogue text
contains any intent, if yes,
executing a process of an entity matching module, if not, terminating the
process, if
impossible to judge, executing a process of a text expanding module;
[0037] the text expanding module, for contextually expanding the dialogue
text, and executing
the process of the entity matching module with respect to the expanded
dialogue text;
[0038] the entity matching module, for identifying a named entity set in the
dialogue text, and
determining any intent knowledge point associated with the named entity set;
[0039] a semantically predicting module, for expressing the dialogue text in
distributed word
vectors, and employing plural pre-trained semantically classifying models for
prediction
to obtain plural pieces of semantic information; and
[0040] a merging and tuning module, for employing an Ensemble framework to
merge and tune
the intent knowledge point and the plural pieces of semantic information, and
obtaining
a user intent.
[0041] Moreover, the device further comprises:
[0042] a text rectifying module, for performing text rectification on the
dialogue text.
4
Date Regue/Date Received 2022-09-23
CA 03176868 2022-09-23
[0043] Further, the text rectifying module is specifically employed for:
[0044] term-segmenting the dialogue text, and identifying any erroneous
segmented term in the
dialogue text;
[0045] obtaining a rectifying term to which the erroneous segmented term
corresponds; and
[0046] replacing the erroneous segmented term in the dialogue text with the
rectifying term.
[0047] Further, the text expanding module is specifically employed for:
[0048] storing user conversation information with one session as a unit;
[0049] associating with contextual information of the dialogue text, and
judging whether the user
intent is changed, wherein the contextual information includes an intent
identifying result
of context of the dialogue text; and
[0050] employing a near-synonym of the context to expand the dialogue text
when the user intent
is not changed.
[0051] Further, the entity matching module is specifically employed for:
[0052] performing a term-segmenting process on the dialogue text according to
a preset
dictionary, and obtaining plural segmented terms;
[0053] matching the plural segmented terms with a preset entity lexicon, and
obtaining the
named entity set; and
[0054] determining an intent knowledge point relevant to the named entity set
from a preset
knowledge base.
[0055] Further, the semantically predicting module is specifically employed
for:
[0056] performing a term-segmenting process on the dialogue text, and
obtaining plural
segmented terms;
[0057] calculating word vectors of the plural segmented terms, and expressing
the word vectors
of the plural segmented terms in distribution; and
[0058] inputting the word vectors of the plural segmented terms expressed in
distribution to the
Date Regue/Date Received 2022-09-23
CA 03176868 2022-09-23
plural semantically classifying models to output the plural pieces of semantic
information.
[0059] Further, the merging and tuning module is specifically employed for:
[0060] determining a final user intent through the Ensemble framework
according to the intent
knowledge point, the plural pieces of semantic information, and preset weights
to which
the plural pieces of semantic information respectively correspond.
[0061] In comparison with prior-art technology, the present invention achieves
the following
advantageous effects:
[0062] 1. When it is impossible to judge whether the dialogue text contains
any intent, intent
information can be complemented in the user dialogue text by contextually
expanding the
dialogue text.
[0063] 2. Semantic association among terms is taken into full consideration
while features are
being extracted by expressing the dialogue text in distributed word vectors
and by deep
feature mining through a deep learning model.
[0064] 3. An Ensemble framework is employed to merge and tune the entity
matching result and
the semantically predicting result to obtain the user intent, whereby are
achieved to more
quickly and accurately identify the user intent, to enhance precision rate in
identification
of the user intent, and to reduce error and incompleteness in identification
of the user
intent, so that guarantee is supplied for the customer service robot to
correctly answer
questions raised by users.
BRIEF DESCRIPTION OF THE DRAWINGS
[0065] To describe the technical solutions in the embodiments of the present
invention more
clearly, drawings required for use in the description of the embodiments will
be briefly
introduced below. Apparently, the drawings introduced below are merely
directed to some
embodiments of the present invention, and it is possible for persons
ordinarily skilled in
the art to base on these drawings to acquire other drawings without creative
effort being
6
Date Regue/Date Received 2022-09-23
CA 03176868 2022-09-23
spent in the process.
[0066] Fig. 1 is a flowchart illustrating an intent identifying method for
application to an
intelligent customer service robot;
[0067] Fig. 2 is a flowchart illustrating specific implementation of step Si
in Fig. 1;
[0068] Fig. 3 is a flowchart illustrating specific implementation of step S3
in Fig. 1;
[0069] Fig. 4 is a flowchart illustrating specific implementation of step S4
in Fig. 1;
[0070] Fig. 5 is a flowchart illustrating specific implementation of step S5
in Fig. 1; and
[0071] Fig. 6 is a block diagram illustrating an intent identifying device for
application to an
intelligent customer service robot.
DETAILED DESCRIPTION OF THE INVENTION
[0072] To make more lucid and clear the objectives, technical solutions and
advantages of the
present invention, the technical solutions in the embodiments of the present
invention will
be clearly and comprehensively described below in conjunction with
accompanying
drawings in the embodiments of the present invention. Apparently, the
embodiments as
described below are merely partial, rather than the entire, embodiments of the
present
invention. All other embodiments makeable by persons ordinarily skilled in the
art on the
basis of the embodiments in the present invention without spending any
creative effort in
the process shall all fall within the protection scope of the present
invention.
[0073] An embodiment of the present invention provides an intent identifying
method for
application to an intelligent customer service robot, the method obtains a
user intent by
7
Date Regue/Date Received 2022-09-23
CA 03176868 2022-09-23
contextually expanding a dialogue text in combination with entity matching
identification
and semantic information prediction, whereby are made possible to more quickly
and
accurately identify the user intent, to enhance precision rate in
identification of the user
intent, and to reduce error and incompleteness in identification of the user
intent, so that
guarantee is supplied for the customer service robot to correctly answer
questions raised
by users.
[0074] Understandably, the method provided by the embodiment of the present
invention is
applicable to any intelligent terminal that includes, but is not limited to, a
table computer,
a personal computer, a smart mobile phone, and a panel computer, etc.
[0075] As should be additionally noted, the terms "first" and "second" etc. as
used in the
description of the present invention are merely meant for descriptive
purposes, rather than
for indicating or implying relative importance. In addition, unless explained
otherwise in
the description of the present invention, the wordings of "plural" and "a
plurality of'
denote the meaning of "two or more".
[0076] Embodiment 1
[0077] This embodiment of the present invention provides an intent identifying
method for
application to an intelligent customer service robot, with reference to what
is shown in
Fig. 1, the method comprises the following steps.
[0078] SO ¨ obtaining a dialogue text of a user.
[0079] The user dialogue can be a speech or a text, when the dialogue is a
speech, the user
dialogue can be converted from speech to text before execution of the
embodiment of the
present invention. Besides, the dialogue text can be a long text, and can also
be a short
text, to which no specific definition is made in the embodiments of the
present invention.
8
Date Regue/Date Received 2022-09-23
CA 03176868 2022-09-23
[0080] Si ¨ performing text rectification on the dialogue text.
[0081] With reference to what is shown in Fig. 2, the specific implementation
process of step Si
can include:
[0082] Sll ¨ term-segmenting the dialogue text, and identifying any erroneous
segmented term
in the dialogue text; and
[0083] 512 ¨ obtaining a rectifying term to which the erroneous segmented term
corresponds,
and replacing the erroneous segmented term in the dialogue text with the
rectifying term.
[0084] Specifically, the rectifying term to which the erroneous segmented term
corresponds can
be obtained on the basis of a dictionary of wrong words, specifically
speaking, with
respect to the erroneous segmented term, a rectification confidence degree to
which each
term in a self-defined standard lexicon corresponds is calculated, and any
term whose
rectification confidence degree is greater than a preset threshold is taken to
serve as the
rectifying term. In addition, it is also possible to employ such modes as edit
distance or
language model to obtain the rectifying term to which the erroneous segmented
term
corresponds, while the specific obtaining process is not specifically defined
in this
embodiment.
[0085] In the above step 512, the rectifying term is mainly used for
rectifying any erroneous
segmented term in the identified text. For example, if an erroneous segmented
term in the
identified text "l-fEelti WM (big treeta application case)" is "A-IE (big
treeta)",
the corresponding rectifying term will be "ttlg (big data)".
[0086] As should be noted, step Si is an optional process.
[0087] In this embodiment, by performing text rectification on the dialogue
text, the dialogue
text with any phrasal error is converted into correct expression that conforms
to field
9
Date Regue/Date Received 2022-09-23
CA 03176868 2022-09-23
logics, so that the user intent can be more accurately identified.
[0088] S2 ¨ judging whether the dialogue text contains any intent, if yes,
executing step S4, if
not, terminating the process, if impossible to judge, executing step S3.
[0089] A dialogue text with intent differs from a dialogue text without intent
relatively greatly
in terms of wording and sentence pattern, it can hence be attempted to
directly employ
some template matching modes to judge whether the user dialogue is a dialogue
with
intent or a dialogue without intent.
[0090] The specific implementation process of judging whether the dialogue
text contains any
intent in step S2 can include:
[0091] searching in the dialogue text whether there is any word group matching
a preset template,
if yes, deciding that the dialogue text contains intent, if not, deciding that
the dialogue
text contains no intent, wherein the preset template can be embodied as a
regular
expression mode.
[0092] Besides, since the text expressed by the user in the customer service
robot may be a
dialogue text of only several terms, this renders the user expression
extremely ambiguous
and unclear, when it is impossible to judge whether the user dialogue contains
any intent
by means of the process of step S2, it is required to contextually expand the
dialogue text.
[0093] S3 ¨ contextually expanding the dialogue text, and executing step S4
after step S3.
[0094] With reference to what is shown in Fig. 3, the specific implementation
process of step S3
can include the following.
[0095] S31 - storing user conversation information with one session as a unit,
associating with
contextual information of the dialogue text, and judging whether the user
intent is
Date Regue/Date Received 2022-09-23
CA 03176868 2022-09-23
changed, wherein the contextual information includes an intent identifying
result of
context of the dialogue text.
[0096] Specifically, with respect to a dialogue text difficult to judge
whether it contains any
intent, it is possible to associate with relevant information of the context
with one session
as a unit, employ conversation information stored in one session to merge
plural dialogue
texts previously input by the user, and judge whether the intent is changed.
[0097] S32 - employing a near-synonym of the context to expand the dialogue
text when the user
intent is not changed.
[0098] Specifically, keywords in the context are extracted to obtain a set of
near-synonyms, and
the set of near-synonyms is used to expand the dialogue text.
[0099] In this embodiment, the dialogue text impossible to judge whether it
contains any intent
is contextually expanded, whereby intent information in the dialogue context
can be
enriched to facilitate subsequent accurate recognition of user intent.
[0100] S4 ¨ identifying a named entity set in the dialogue text, and
determining any intent
knowledge point associated with the named entity set.
[0101] With reference to what is shown in Fig. 4, the specific implementation
process of step S4
can include the following.
[0102] S41 - performing a term-segmenting process on the dialogue text
according to a preset
dictionary, and obtaining plural segmented terms.
[0103] Specifically, a preset term-segmenting mode is employed to perform a
term-segmenting
process on the dialogue text according to a preset dictionary to obtain a
plurality of
11
Date Regue/Date Received 2022-09-23
CA 03176868 2022-09-23
characters or character sequences, and characters or character strings with
practical
semantics are screened out of the character sequences as obtained to serve as
a term-
segmenting result. The preset term-segmenting mode can be to a term-segmenting
mode
that is based on character matching, based on sematic understanding, or based
on statistics.
[0104] S42 - matching the plural segmented terms with a preset entity lexicon,
and obtaining the
named entity set.
[0105] Specifically, with respect to each segmented term in plural segmented
terms, a matching
degree between each named entity in the entity lexicon and this segmented term
is
calculated, and any named entity whose matching degree is greater than a
preset threshold
is taken to serve as the named entity matching this segmented term. In
addition, it is
possible to employ similarities based on Hamming distance to calculate the
matching
degree between each named entity in the entity lexicon and this segmented
term.
[0106] For example, with respect to segmented terms "Shanghai" and "aged 60"
in a user
dialogue text, a named entity "region" of "Shanghai" and a named entity "age"
of "aged
60" can be matched and obtained from the entity lexicon.
[0107] S43 - determining an intent knowledge point relevant to the named
entity set from a preset
knowledge base.
[0108] In this embodiment, plural entities correspond to one intent knowledge
point, and the
intent knowledge point is used to indicate intent information; normalized
intent
knowledge points can be collected and sorted in advance according to
historically
accumulated chat data carried out between customer service and users,
corresponding
plural entities are then determined for each intent knowledge point, and
preliminary
prediction of the user intent can be obtained by matching with the entity
lexicon.
12
Date Regue/Date Received 2022-09-23
CA 03176868 2022-09-23
[0109] Specifically, relevancy between the named entity set and each intent
knowledge point in
the knowledge base is calculated, and any intent knowledge point in the
knowledge base
associated with the named entity set is determined.
[0110] S5 ¨ expressing the dialogue text in distributed word vectors, and
employing plural pre-
trained semantically classifying models for prediction to obtain plural pieces
of semantic
information.
[0111] With reference to what is shown in Fig. 5, the specific implementation
process of step S5
can include the following.
[0112] S51 - performing a term-segmenting process on the dialogue text, and
obtaining plural
segmented terms.
[0113] Specifically, the specific process of this step is identical with step
S41, so it is not
redundantly described here.
[0114] S52 - calculating word vectors of the plural segmented terms, and
expressing the word
vectors of the plural segmented terms in distribution.
[0115] Specifically, word vectors to which term units correspond can be
obtained through a
Word2Vec model, and the word vectors are expressed in distribution.
[0116] Word2Vec is a specific means of word embedding of natural language
processing (NLP),
it can characterize semantic information of terms in the mode of word vectors
through
the learning of texts, i.e., semantically close words are arranged very close
in distance
within an embedded space through this embedded (low-dimensional) space.
[0117] S53 - inputting the word vectors of the plural segmented terms
expressed in distribution
13
Date Regue/Date Received 2022-09-23
CA 03176868 2022-09-23
to the plural semantically classifying models to output the plural pieces of
semantic
information.
[0118] The process of training the plural semantically classifying models in
step S5 includes the
following.
[0119] Q&A data is obtained from the database, the Q&A data is preprocessed,
and the
preprocessed Q&A data is marked.
[0120] The Q&A data includes information of Q&A pairs accumulated by field
human customer
service in answering questions raised by users.
[0121] Specifically, the Q&A data can be preprocessed by means of keyword
extraction and
template rules, partial data without intent is filtered away, and the
preprocessed Q&A data
is semantically marked by marking personnel.
[0122] For instance, semantic classifications within the field can be
particularized into variegated
classifications including telephone charges, gift cards, managements of money
matters,
and Change Treasure, etc., and the Q&A data is marked in advance by marking
personnel.
[0123] The marked Q&A data is divided into a training set and a testing set by
the mode of offline
pretraining.
[0124] Q&A statements in the training set are expressed by distribution of
word vectors and are
trained in a deep neural network, the testing set is used to test the trained
deep neural
network, and semantically classifying models whose prediction precisions
satisfy a
precision threshold are constructed.
[0125] The plural semantically classifying models can be embodied as such
various deep-
14
Date Regue/Date Received 2022-09-23
CA 03176868 2022-09-23
learning semantically classifying models as TextCNN, RNN, LSTM, and CAPsNet,
etc.,
as can be understood by persons skilled in the art, the model training
strategy can be a
conventional strategy of the corresponding network, to which no explanation is
made in
this context.
[0126] After deep neural networks have been trained by means of the training
set, the testing set
can be used to respectively test the plural trained deep neural networks, to
appraise
prediction precision rates of the deep neural networks, and to adjust network
parameters
of the deep neural networks in accordance with model prediction precision
rate, so as to
construct the semantically classifying models whose prediction precisions
satisfy the
precision threshold.
[0127] In this embodiment of the present invention, semantic association among
terms is taken
into full consideration while features are being extracted by expressing Q&A
data well
marked with semantic classifications by distribution of word vectors and by
deep feature
mining through deep learning models, to obtain the semantically classifying
models. It is
hence possible to use the plural semantically classifying models to quickly
and accurately
predict semantic information with respect to the user dialogue text expressed
by
distribution of word vectors.
[0128] S6 ¨ employing an Ensemble framework to merge and tune the intent
knowledge point
and the plural pieces of semantic information, and obtaining a user intent.
[0129] Specifically, the final user intent is determined through the Ensemble
framework
according to the intent knowledge point, the plural pieces of semantic
information, and
preset weights to which the plural pieces of semantic information respectively
correspond.
[0130] The basic conception of the Ensemble framework is to form a strong
classification
framework by taking full advantages of different classification algorithms and
Date Regue/Date Received 2022-09-23
CA 03176868 2022-09-23
compensating one's disadvantage by another's advantage. Plural classifiers are
merged
together to realize the optimum combination.
[0131] In the intent identifying method for application to an intelligent
customer service robot
provided by the embodiments of the present invention, when it is impossible to
judge
whether the dialogue text contains any intent, intent information can be
complemented in
the user dialogue text by contextually expanding the dialogue text; semantic
association
among terms is taken into full consideration while features are being
extracted by
expressing the dialogue text in distributed word vectors and by deep feature
mining
through a deep learning model; an Ensemble framework is employed to merge and
tune
the entity matching result and the semantically predicting result to obtain
the user intent,
whereby are achieved to more quickly and accurately identify the user intent,
to enhance
precision rate in identification of the user intent, and to reduce error and
incompleteness
in identification of the user intent, so that guarantee is supplied for the
customer service
robot to correctly answer questions raised by users.
[0132] Embodiment 2
[0133] As realization of the intent identifying method for application to an
intelligent customer
service robot in Embodiment 1, this embodiment of the present invention
provides an
intent identifying device for application to an intelligent customer service
robot, with
reference to what is shown in Fig. 6, the device comprises:
[0134] a text obtaining module 60, for obtaining a dialogue text of a user;
[0135] an intent judging module 62, for judging whether the dialogue text
contains any intent, if
yes, executing a process of an entity matching module, if not, terminating the
process, if
impossible to judge, executing a process of a text expanding module 63;
[0136] the text expanding module 63, for contextually expanding the dialogue
text, and
executing the process of the entity matching module with respect to the
expanded
dialogue text;
16
Date Regue/Date Received 2022-09-23
CA 03176868 2022-09-23
[0137] the entity matching module 64, for identifying a named entity set in
the dialogue text, and
determining any intent knowledge point associated with the named entity set;
[0138] a semantically predicting module 65, for expressing the dialogue text
in distributed word
vectors, and employing plural pre-trained semantically classifying models for
prediction
to obtain plural pieces of semantic information; and
[0139] a merging and tuning module 66, for employing an Ensemble framework to
merge and
tune the intent knowledge point and the plural pieces of semantic information,
and
obtaining a user intent.
[0140] Moreover, the device further comprises:
[0141] a text rectifying module 61, for performing text rectification on the
dialogue text.
[0142] Further, the text rectifying module 61 is specifically employed for:
[0143] term-segmenting the dialogue text, and identifying any erroneous
segmented term in the
dialogue text;
[0144] obtaining a rectifying term to which the erroneous segmented term
corresponds; and
[0145] replacing the erroneous segmented term in the dialogue text with the
rectifying term.
[0146] Further, the text expanding module 63 is specifically employed for:
[0147] storing user conversation information with one session as a unit;
[0148] associating with contextual information of the dialogue text, and
judging whether the user
intent is changed, wherein the contextual information includes an intent
identifying result
of context of the dialogue text; and
[0149] employing a near-synonym of the context to expand the dialogue text
when the user intent
is not changed.
[0150] Further, the entity matching module 64 is specifically employed for:
[0151] performing a term-segmenting process on the dialogue text according to
a preset
dictionary, and obtaining plural segmented terms;
17
Date Regue/Date Received 2022-09-23
CA 03176868 2022-09-23
[0152] matching the plural segmented terms with a preset entity lexicon, and
obtaining the
named entity set; and
[0153] determining an intent knowledge point relevant to the named entity set
from a preset
knowledge base.
[0154] Further, the semantically predicting module 65 is specifically employed
for:
[0155] performing a term-segmenting process on the dialogue text, and
obtaining plural
segmented terms;
[0156] calculating word vectors of the plural segmented terms, and expressing
the word vectors
of the plural segmented terms in distribution; and
[0157] inputting the word vectors of the plural segmented terms expressed in
distribution to the
plural semantically classifying models to output the plural pieces of semantic
information.
[0158] Further, the merging and tuning module 66 is specifically employed for:
[0159] determining a final user intent through the Ensemble framework
according to the intent
knowledge point, the plural pieces of semantic information, and preset weights
to which
the plural pieces of semantic information respectively correspond.
[0160] The intent identifying device for application to an intelligent
customer service robot
provided by this embodiment pertains to the same inventive concept as the
intent
identifying method for application to an intelligent customer service robot
provided by
the foregoing embodiment of the present invention, can execute the intent
identifying
method for application to an intelligent customer service robot provided by
any
embodiment of the present invention, and possesses functional modules and
achieves
advantageous effects to which the intent identifying method for application to
an
intelligent customer service robot corresponds. Technical details not
particularized in this
embodiment can be inferred from the intent identifying method for application
to an
intelligent customer service robot provided by the foregoing embodiment of the
present
invention, and are not redundantly described in this context.
18
Date Regue/Date Received 2022-09-23
CA 03176868 2022-09-23
[0161] All the above optional technical solutions can be randomly combined to
form optional
embodiments of the present invention, and these are not redundantly described
on a one-
by-one basis.
[0162] As understandable by persons ordinarily skilled in the art, realization
of the entire or
partial steps of the aforementioned embodiments can be completed by hardware,
or by a
program instructing relevant hardware, the program can be stored in a computer-
readable
storage medium, and the storage medium can be a read-only memory, a magnetic
disk, or
an optical disk, etc.
[0163] What is described above is merely directed to preferred embodiments of
the present
invention, and is not meant to restrict the present invention. Any
modification, equivalent
substitution, and improvement makeable within the spirit and principle of the
present
invention shall all be covered by the protection scope of the present
invention.
19
Date Regue/Date Received 2022-09-23
