JP6526607B2 - Learning apparatus, learning method, and learning program - Google Patents

Description

The present invention, a learning device, a learning method, about your and learning programs.

  2. Description of the Related Art Conventionally, techniques for labeling information using techniques called Recurrent Neural Network (RNN) and deep learning are known. As an example of such a technique, there is known a technique of learning a learning device to output a label indicating the content or type of a sentence according to the content of an input sentence and applying a label to the sentence .

“Learning phrase representation using RNN encoder-decoder for statistical machine translation.” K. Cho, B. van Merrienboer, C. Gulcehre, D. Bahdanau, F. Bougares, H. Schwenk, and Y. Bengio. In Proceedings of the 2014 Confer-ence on EMNLP, 2014.

So ,, Perelygin, A .; Wu, J .; Chuang, J .; Manning, CD; Ng, A .; and Potts, C. 2013. Recursive deep models for semantic compositionality over a sentiment treebank. The 2013 Conference on Empirical Methods in Natural Language Processing (EMNLP 2013), 1631-1642. Association for Computational Linguistics.

  However, in the prior art, there are cases where the learning device can not provide appropriate labels.

  For example, when learning a learning device to output a label indicating the content or type of a sentence according to the content of the input sentence, the relationship between the sentence and the label to be attached to the sentence is learned Will make the vessel learn. However, if correct data including the sentence and the label to be attached to the sentence can not be prepared sufficiently, the relationship between the sentence and the label can not be properly learned, and the learning device assigns the appropriate label. There is a risk that you will not be able to

  This application is made in view of the above, and it aims at improving the accuracy of the label given to a sentence.

  A learning apparatus according to the present application includes a first learning unit that causes a learning device and a restoring device included in a first recurrent neural network to learn a relationship between an utterance and a response in a predetermined dialog; A first generation unit for generating a second recurrent neural network using at least one of the above-described learners or reconstructors, and a relationship between a sentence and a label given to the sentence, the second recurrent And a second learning unit for making the neural network learn.

  According to one aspect of the embodiment, the accuracy of the label given to the text can be improved.

FIG. 1 is a diagram illustrating an example of learning processing performed by the learning device according to the embodiment. FIG. 2 is a diagram illustrating an exemplary configuration of a learning device according to the embodiment. FIG. 3 is a diagram showing an example of information registered in the dialogue data database according to the embodiment. FIG. 4 is a diagram showing an example of information registered in a learning data database according to the embodiment. FIG. 5 is a first diagram showing a variation of RNN generated by the learning device according to the embodiment. FIG. 6 is a second diagram showing a variation of the RNN generated by the learning device according to the embodiment. FIG. 7 is a flowchart illustrating an example of the flow of learning processing according to the embodiment. FIG. 8 is a diagram illustrating an example of the hardware configuration.

Hereinafter, a learning apparatus, a learning method, program parameters, and a mode for implementing a learning program according to the present application (hereinafter, referred to as “embodiment”) will be described in detail with reference to the drawings. Note that the learning device, the learning method, the program parameters, and the learning program according to the present application are not limited by the embodiment. Moreover, the same code | symbol is attached | subjected to the same site | part in the following each embodiment, and the overlapping description is abbreviate | omitted.
[Embodiment]

[1-1. Example of learning device]
First, an example of the learning process performed by the learning device will be described with reference to FIG. FIG. 1 is a diagram illustrating an example of learning processing performed by the learning device according to the embodiment. In FIG. 1, the learning device 10 is an information processing device that executes learning processing described below, and is realized by, for example, a server device, a cloud system, or the like.

  More specifically, the learning device 10 can communicate with any device such as the terminal device 100 used by the user via a predetermined network N such as the Internet. In addition, the learning device 10 can communicate with an input device 200 (see, for example, FIG. 2) for acquiring an interaction between arbitrary users.

  The terminal device 100 and the input device 200 are realized by a smart device such as a smartphone or a tablet, an information processing device such as a server device such as a desktop PC (Personal Computer) or a notebook PC. The learning device 10 is realized by any information processing device such as a server device or a cloud system.

[1-2. About learning process]
Here, the learning device 10 learns a model for adding attribute information indicating an attribute such as the type and content of the sentence to the input sentence. Such attribute information is called a label, a tag, or metadata, and is used when searching for or selecting a sentence. In the following description, attribute information is referred to as a label.

  For example, when a plurality of words included in a sentence are input in the order in which the words appear in the sentence, the learning device 10 learns the features of each word and the features of the order in which the words appear The network (hereinafter referred to as “recurrent neural network (RNN)) is learned. Here, when a plurality of pieces of information are input in order, the RNN codes the features of each piece of information and the features of the input order It is a neural network having a learning unit (encoder) that outputs the encoded feature, and a restoring unit (decoder) that restores new information based on the feature output from the learning unit. A process is performed to restore a plurality of pieces of information input to the learning device in the order in which they are input, from the features output from the learning device.

  For example, the learning device 10 acquires, as learning data used for RNN learning, a set of a predetermined sentence and a label to be attached to the sentence. In such a case, when a word contained in a sentence is input to the RNN, the learning device 10 learns the RNN so as to output a label to be attached to the sentence. For example, the learning device 10 performs RNN learning using a back propagation method or any learning method used for learning such as deep learning or neural network.

  However, when the number of learning data can not be acquired sufficiently, the feature of each word included in the sentence, the feature of the order in which the words appear, etc. can not be learned appropriately, and the label attached to the sentence Accuracy may decrease.

  Therefore, the learning device 10 executes the following learning process. First, the learning device 10 causes the learning device and the restoring device of the first RNN to learn the relationship between the utterance and the response in the predetermined dialogue. Subsequently, the learning device 10 generates a second RNN using at least one of the learning device and the restoring device in which the learning has been performed. Then, the learning device 10 causes the second RNN to learn the relation between the sentence and the label given to the sentence. That is, the learning device 10 performs pre-training on the relationship between the sentence and the label using the utterance and the response included in the dialogue.

  Here, it is predicted that the utterance and the response included in the dialogue include hidden commonality. For example, when there is an animation or movie based on a novel, the utterance and the response included in the dialogue may be a dialogue about a novel, a dialogue about an animation, or a dialogue about a movie, etc. There is a hidden commonality. In addition, it is predicted that the utterance included in the dialogue and the response to the utterance have a common context (context). Such commonality is predicted to be similar to commonality between the context of sentences and the context of labels. Then, if the relationship between the speech and the response is learned in advance, it is considered that the commonality and co-occurrence of the text and the label can be reinforced.

  Therefore, the learning device 10 performs pre-training on the learning device and the restoring device included in the RNN using the utterance and the response. For example, when a speech is input to a learning device, the learning device 10 performs pre-training on the learning device and the restoring device such that the restoration device outputs a response to the speech. More specifically, the learning device 10 causes the learning device to learn the features that each word has, together with the order in which the words included in the utterance appear in the utterance. Then, the learning apparatus 10 restores the decompressor so that each word included in the response is restored from the feature of the speech encoded by the learning device, that is, the information output from the learning device, along with the order in which each word appears in the response. To learn.

  In addition, if the learning device 10 associates the speech and the response with either the learning device or the restoration device, it is necessary to cause the speech to be learned in association with the learning device, and to make the response correspond to the restoration device and learned. There is no. For example, the learning device 10 may associate the response with the learning device to learn, and associate the utterance with the recovery device to learn. Specifically, the learning device 10 may perform learning of the learning device and the restoring device such that the restoring device outputs an utterance corresponding to the response when the response is input to the learning device. For example, the learning device 10 causes the learning device to learn the features that each word has with the order in which the words included in the response appear in the response. Then, the learning device 10 may make the decompressor learn so as to restore each word included in the utterance along with the order in which each word appears in the utterance from the feature encoded by the learning device. Further, the learning device 10 may perform learning of a model serving as a learning / restoring device by sharing parameters between the learning device and the restoration device. For example, the learning device 10 may perform learning of one RNN operating as a learning and restoring device.

[1-3. About generation of RNN]
Here, the learning device 10 may learn the relationship between the sentence and the label by using the pre-trained RNN, that is, the first RNN as it is. Also, even if the learning device 10 generates a new RNN using at least one of a learning device or a restoring device included in the first RNN, and causes the generated RNN to learn the relationship between a sentence and a label. Good.

  For example, the learning device 10 generates a second RNN using the learning device included in the first RNN and the newly generated restorer. Here, if the newly generated restorer is the restorer included in the first RNN, any restorer is applicable, and, for example, the transfer coefficient at the time of transmitting information between neurons is initialized ( For example, it may be a randomizer). It is considered that the feature of the hidden commonness included in the speech and the response is reflected on the learning device side in the second RNN generated in this manner. In other words, it is considered that the second RNN is likely to converge the transmission coefficients between the respective neurons so as to output an appropriate label when the sentence is input. Therefore, even if the amount of learning data is not sufficient, the second RNN is related to the relation between the sentence and the label included in the learning data, that is, the feature of the hidden commonality possessed by the sentence and the label. It is thought that you can learn quickly.

  Therefore, when a sentence of learning data is input to the second RNN, the learning device 10 performs learning of decoding of the second RNN so that the second RNN outputs a label to be attached to the input sentence. For example, the learning device 10 connects a randomly initialized restorer to the first RNN's learner, fixes the parameters of the learner, and updates the restorer-side parameters. As a result, since the learning device 10 can generate an RNN that gives an appropriate label to a sentence, the accuracy of the label given to the sentence can be improved.

  Here, the learning device 10 may use the second RNN as it is for learning in the third RNN, that is, when applying a label to a sentence other than the learning data, using the second RNN. A new third RNN may be generated, and the new third RNN may be used as the RNN used in the measurement (hereinafter referred to as “measurement RNN”). For example, it is predicted that the feature of the relation between the sentence and the label is learned in the decompressor included in the second RNN. More specifically, the decompressor included in the second RNN is added to the sentence in which the plurality of words appear in the order, from the information indicating the features of the plurality of words and the features of the order in which each word appears Learning is performed so as to output information indicating a label to be done. Therefore, the learning device 10 may use, as the measurement RNN, the third RNN further subjected to the entire learning with respect to the third RNN including the restoration device included in the second RNN and the learning device included in the first RNN. Further, the learning device 10 may use the first RNN's learning device as the measurement RNN without passing through the second RNN, and may use the third RNN subjected to the overall learning. Also, at the time of learning of the third RNN, the learning device 10 fixes the parameters of the learning device and updates only the parameters on the restoration device side, and removes the fixing of the learning device parameters again after some learning has progressed. The entire third RNN may be trained (training).

  Then, the learning device 10 may assign a label to a sentence to which a label is to be assigned, using the third RNN. For example, the learning device 10 inputs each word included in the sentence in the order in which each word appears in the input layer of the learning device included in the third RNN, and the information output from the output layer of the learning device, ie, the input Feature information indicating features of each word included in the sentence and features of the order in which each word appears is input to the input layer of the decompressor. Then, the learning device 10 determines the label to be attached to the text from the information indicating the label output by the decompressor based on the feature information.

[1-4. About variations of RNN]
Here, the learning device 10 may generate the measurement RNN using any process, as long as it uses a learner or a restorer that has performed pre-training using the utterance and the response included in the dialogue. . For example, the learning device 10 generates a second RNN from the learning device included in the first RNN and a new decompressor, and learns the second RNN using the learning data. Then, the learning device 10 may use the second RNN subjected to the learning as the measurement RNN as it is.

  Also, the learning device 10 generates a second RNN using the learning device included in the first RNN and a new recovery device, and learns the second RNN using learning data. Subsequently, the learning device 10 generates a third RNN using the restoration device included in the second RNN and the learning device included in the first RNN, and performs learning of the third RNN using learning data. Then, the learning device 10 may output the third RNN learned using the learning data as the measurement RNN.

  The learning device 10 may use the same learning data or may use different learning data when performing learning of the second RNN and when performing learning of the third RNN. In addition to learning data which is a set of a sentence and a label to be attached to the sentence, the learning device 10 sets a pair of an utterance and a response included in the dialogue as learning data, and learns the second and third RNNs. You may go.

[1-5. About sentences and labels]
Here, the learning device 10 may perform learning of the RNN to which an arbitrary label given to an arbitrary sentence is given. That is, the sentence to which the learning device 10 applies a label may be a sentence related to an article such as a novel, a newspaper article, a blog, a microblog, a critique, a post, or a record such as a medical record. The sentence may be a single sentence or may be a sentence consisting of a plurality of sentences. For example, the learning device 10 may learn sentences and the like in units of books or works.

  Further, the label given by the learning device 10 may be a word or a sentence indicating the content or attribute of the sentence, or may be a numerical value or a character string used for a search or the like.

[1-6. Example of formula]
Next, an example of a mathematical expression indicating the above-described learning process will be described. For example, if the state of the “j” -th node of the learning device at time “t” is described as “hj <t>”, the state of each node in the learning device can be expressed by the following equation (1). Here, the waved “hj <t>” in the formula (1) can be expressed by the formula (2), and the “zj” and the “rj” can be expressed by the following formulas (3) and (4) expressed. Here, “xt” indicates the “t” -th input. Moreover, "W" and "U" of Formula (2)-Formula (4) are parameters which show the connection coefficient of a learning device.

  Then, the learning result by the learning device can be expressed by equation (5).

  Here, the output of the learner is transmitted to each node of the restorer. The input to such a decompressor node can be shown by equation (6).

  Then, assuming that the state of the “j” -th node of the decompressor at time “t” is “h′j <t>”, “h′j <t>” can be expressed by the following equation (7) it can. Here, the waved “h′j <t>” in the formula (7) can be expressed by the following formula (8), and “z′j” and “r′j” are 9) and equation (10). Moreover, "W '" and "U'" of Formula (8)-Formula (10) are parameters which show the connection coefficient of a decompressor.

  For example, in the learning device 10, the learning device learns a plurality of words included in the sentence, that is, the features and order of the word group, and the restoring device learns the plurality of words input to the learning device from the output of the learning device. "W", "U", "W '", and "U" in equations (1) to (10) so as to output information indicating labels to be attached to the sentences that appear in the order in which they are input to Learn the value of ''.

[1-7. Example of learning process]
Next, an example of the learning process performed by the learning device 10 will be described with reference to FIG. First, the learning device 10 accepts a dialogue such as the user's statement A or statement B as an input (step S1). For example, the input device 200 acquires the voice of the dialog performed between a plurality of users using a voice acquisition device that acquires voice such as a microphone. Then, the input device 200 converts the dialogue into text data using any speech recognition technology, and transmits the converted text data, that is, the dialogue text data to the learning device 10.

  More specifically, the input device 200 sets text data of a speech A by a certain user as speech data (hereinafter sometimes referred to as "speech data"), and the response of the other user to the speech A The text data of statement B which is is generated as data of the response (which may hereinafter be described as "response data"). Then, the input device 200 transmits the speech data and the response data to the learning device 10 as data of dialogue.

  In such a case, the learning device 10 associates the utterance and the response included in the input dialogue text with either the learning device or the recovery device of the first RNN, respectively, and hides the hidden information inherent in the utterance and the response. That is, pre-training the hidden commonality inherent in speech and response (step S2). For example, the learning device 10 extracts an utterance and a response corresponding to the utterance from text data of a dialog. Then, when the learning device 10 inputs each word included in the utterance to the learning device EN01 included in the first RNN in the order included in the utterance, the utterance input to the learning device EN01 from the output of the learning device EN01 The learner EN01 and the restorer DE01 are trained such that the restorer DE01 outputs the words included in the response corresponding to the order in which the words are included in the response. More specifically, the learning device 10 corrects the transfer coefficients of the learning device EN01 and the restoring device DE01.

  Next, the learning device 10 generates a second RNN combining the pre-trained learning device EN01 and the new restoring device DE02, and from the sentence of the labeled sentence (that is, learning data), the labeled sentence The second RNN is learned to generate a label (step S3). For example, when the learning device 10 inputs each word included in the sentence of the labeled sentence to the learning device EN01 included in the second RNN in the order included in the sentence, the output of the learning device EN01 indicates the labeled sentence The learner EN01 and the restorer DE02 are trained so that the restorer DE02 outputs information indicating the label of.

  Here, it is assumed that the learning device 10 has acquired a sentence to which a label is to be attached from the terminal device 100 (step S4). In such a case, the learning device 10 executes processing at the time of measurement. More specifically, the learning device 10 uses, as a measurement RNN, a third RNN in which the restorer DE02 that has performed label learning using a labeled sentence and the learning device EN01 that has performed pre-training are terminal devices. The label given to the sentence acquired from 100 is determined (step S5). For example, the learning device 10 inputs the words included in the sentence acquired from the terminal device 100 to the learning device EN01 of the measurement RNN in the order in which the words appear. Then, the learning device 10 determines a label to be attached to the text based on the information output from the restoration device DE02 of the measurement RNN. Thereafter, the learning device 10 outputs information indicating the label that is the determination result to the terminal device 100 (step S6).

[2. Configuration of learning device]
Hereinafter, an example of a functional configuration of the learning device 10 for realizing the above-described learning process will be described. FIG. 2 is a diagram illustrating an exemplary configuration of a learning device according to the embodiment. As shown in FIG. 2, the learning device 10 includes a communication unit 20, a storage unit 30, and a control unit 40.

  The communication unit 20 is realized by, for example, a network interface card (NIC). The communication unit 20 is connected to the network N in a wired or wireless manner, and transmits and receives information to and from the terminal device 100 and the input device 200.

  The storage unit 30 is realized by, for example, a semiconductor memory device such as a random access memory (RAM) or a flash memory, or a storage device such as a hard disk or an optical disk. The storage unit 30 also stores a dialogue data database 31 and a learning data database 32. Note that the storage unit 30 further stores, for example, information indicating the relationship between the node of the RNN to be learned and the transfer coefficient.

  In the dialogue data database 31, dialogue data used for pre-training is registered. For example, FIG. 3 is a diagram showing an example of information registered in the dialogue data database according to the embodiment. In the example shown in FIG. 3, information having items such as “dialogue ID”, “speech information”, “response information”, etc. is registered in the dialogue data database 31. In the dialogue data database 31, information other than the information shown in FIG. 3 may be registered, such as the attributes of the user who made the utterance or the response.

  Here, "dialogue ID (Identifier)" is information for identifying a dialogue. Also, "utterance information" is text data of an utterance. Also, “response information” is text data of a response to the associated utterance. In the example shown in FIG. 3, conceptual values such as “dialogue # 1”, “utterance # 1”, “response # 1”, etc. are described, but in practice, an identifier for identifying dialogue, text data, etc. Etc. are registered in the dialogue data database 31.

  For example, in the example illustrated in FIG. 3, the dialog ID “dialogue # 1”, the utterance information “utterance # 1”, and the response information “response # 1” are registered in association with each other. Such information indicates that the text data of the utterance included in the dialogue indicated by the dialogue ID "dialogue # 1" is "utterance # 1" and the text data of the response to this utterance is "response # 1". .

  Return to FIG. 2 and continue the description. In the learning data database 32, learning data used for learning of the RNN, that is, sentences with labels are registered. For example, FIG. 4 is a diagram showing an example of information registered in a learning data database according to the embodiment. In the example shown in FIG. 4, information having items such as “learning data ID”, “text information”, “label information” and the like is registered in the learning data database 32. In addition to the information shown in FIG. 3, arbitrary information related to the labeled text may be registered in the learning data database 32.

  Here, “learning data ID” is information for identifying a dialogue as learning data. Also, “text information” is text of labeled text, that is, text data. Further, “label information” is information indicating a label to be provided to the associated text. In the example shown in FIG. 4, conceptual values such as “data # 1”, “text # 1”, “label # 1”, etc. are described, but actually, an identifier for identifying a text with a label, Text data and the like are registered in the learning data database 32.

  For example, in the example illustrated in FIG. 4, the learning data ID “data # 1”, the text information “text # 1”, and the label information “label # 1” are registered in association with each other. Such information is that the text data of the sentence included in the labeled sentence indicated by the learning data ID "data # 1" is "sentence # 1", and the label to be assigned to this sentence is "label # 1. It shows that it is ".

  Return to FIG. 2 and continue the description. The control unit 40 is a controller, for example, a processor such as a central processing unit (CPU) or a micro processing unit (MPU), and executes various programs stored in the storage device in the learning device 10 by using a RAM or the like. It is realized by being implemented as a work area. The control unit 40 is a controller, and may be realized by an integrated circuit such as an application specific integrated circuit (ASIC) or a field programmable gate array (FPGA).

  As shown in FIG. 2, the control unit 40 includes an acquisition unit 41, a first learning unit 42, a second learning unit 43, a generation unit 44, and a measurement unit 45.

  The acquisition unit 41 acquires an utterance in a predetermined dialogue and a response to the utterance. For example, when acquiring the text data of dialog from the input device 200, the acquiring unit 41 analyzes the acquired text data, and extracts the text data of the utterance and the text data of the response to the utterance. Then, the acquisition unit 41 registers the extracted text data of the utterance and the text data of the response in the dialog data database 31 as dialog data. The acquiring unit 41 may extract an utterance and a response corresponding to the utterance using any character string analysis technology such as morphological analysis.

  The first learning unit 42 causes the learning device and the restoring device of the first RNN to learn the relationship between an utterance and a response in a predetermined dialogue. That is, the first learning unit 42 performs pre-training of the RNN using the dialogue.

  For example, the first learning unit 42 acquires the utterance and the response included in the dialogue from the dialogue data database 31. In addition, the first learning unit 42 prepares a first RNN to be subjected to pre-training, such as the initialized first RNN. Then, when the first learning unit 42 inputs an utterance to the first RNN learning unit, the first learning unit 42 performs learning of the learning unit and the restoring unit so that the first RNN restoration unit outputs a response to the utterance. More specifically, the first learning unit 42 inputs each word included in the speech to the learning device in the order of appearance in the speech, and each word included in the response is extracted from the features encoded by the learning device. Train the learner and restorer to restore the words with the order in which they appear in the response. That is, the first learning unit 42 causes the learning device and the restoring device to learn the features of each word of the utterance and the features of the order in which the words appear.

  The first learning unit 42 may perform learning of the learning device and the restoring device so that when the response is input to the learning device, the restoring device outputs an utterance corresponding to the response. For example, the first learning unit 42 causes the learning device to learn the features possessed by each word along with the order in which the words included in the response appear in the utterance, and the features encoded by the learning device The decompressor may be trained to recover the words with the order in which each word appears in the response.

  The second learning unit 43 uses a sentence with a label, which is learning data registered in the learning data database 32, to generate a second RNN generated using at least one of a learning device or a restoring device included in the first RNN. And, do learning of the 3rd RNN. An example of processing performed by the second learning unit 43 will be described later.

  The generating unit 44 generates the second RNN using at least one of the learning device or the restoring device that has been learned by the first learning unit 42, that is, the learning device or the restoring device included in the first RNN. For example, the generation unit 44 generates a second RNN using the learning device included in the first RNN and the new recovery device.

  When such a second RNN is generated, the second learning unit 43 causes the second RNN to learn the relationship between the sentence of the labeled sentence registered in the learning data database 32 and the label of the labeled sentence.

  In addition, when the second learning unit 43 performs the second RNN learning, the generating unit 44 performs the learning by the first RNN learning device for which the first learning unit 42 has performed the pre-training, and the second learning unit 43 A third RNN is generated that includes the performed second RNN's restorer. Then, the generation unit 44 registers the generated third RNN as a measurement RNN in the storage unit 30 or the like.

  The second learning unit 43 may perform further learning using learning data for the third RNN. For example, the second learning unit 43 causes the third RNN to learn the relationship between the sentence of the labeled sentence registered in the learning data database 32 and the label of the labeled sentence. Then, the second learning unit 43 may register the third RNN subjected to the learning in the storage unit 30 or the like as the measurement RNN.

  The measurement unit 45 executes measurement processing using the measurement RNN. For example, when the measuring unit 45 receives a sentence to which a label is to be attached from the terminal device 100, the third RNN generated by the generating unit 44 is specified as the measuring RNN, and the label given to the received sentence is specified. Do. For example, the measuring unit 45 inputs each word included in the received sentence into the learning unit of the measurement RNN in the order of appearance in the sentence, and receives it based on the information output by the restoration unit of the measurement RNN. Identify the label given to the sentence Then, the measuring unit 45 transmits information indicating the identified label to the terminal device 100.

  The measuring unit 45 may specify the label given to the received sentence as the third RNN for which the second learning unit 43 further performs learning using the sentence with label as the measurement RNN. In addition, the measurement unit 45 may specify the label given to the received sentence, using the second RNN learned by the second learning unit 43 as the measurement RNN.

[3. Variation of RNN]
Next, variations of generation of RNN used by the learning device 10 as a measurement RNN will be described with reference to FIGS. 5 and 6. FIG. 5 is a first diagram showing a variation of RNN generated by the learning device according to the embodiment. FIG. 6 is a second diagram showing a variation of the RNN generated by the learning device according to the embodiment.

  First, an example of processing for setting the second RNN as a measurement RNN will be described using FIG. 5. For example, the learning device 10 generates a first RNN including a learning device EN01 and a restoring device DE01, and performs pre-training using an utterance and a response (step S10). Subsequently, the learning device 10 takes out the learning device EN01 included in the first RNN (step S11), generates a second RNN using the learning device EN01 and the new decompressor DE02, and generates a sentence with a label and a label The second RNN is learned by using (step S12). Then, the learning apparatus 10 uses the restoration device DE02 of the second RNN that has performed learning (step S13), and the learning device EN01 of the second RNN that has performed learning (step S14) as the measurement RNN. To execute a measurement process for identifying the label (step S15).

  Next, with reference to FIG. 6, an example of processing for setting the third RNN subjected to learning using a labeled sentence as the measurement RNN will be described. For example, the learning device 10 generates a first RNN including a learning device EN01 and a restoring device DE01, and performs pre-training using an utterance and a response (step S20). Subsequently, the learning device 10 takes out the learning device EN01 included in the first RNN (step S21), generates a second RNN using the learning device EN01 and the new decompressor DE02, and generates a sentence with a label and a label The second RNN is learned by using (step S22).

  Also, the learning device 10 takes out the restorer DE02 of the second RNN which has performed learning (step S23), and generates a third RNN combined with the learning device EN01 included in the first RNN which has performed pre-training (step S24). The learning of the third RNN is executed again using the labeled text (step S25). Then, the learning apparatus 10 uses the restoration device DE02 of the third RNN that has performed learning (step S26), and the learning device EN01 of the third RNN that has performed learning (step S27) as the measurement RNN. To execute a measurement process for identifying the label (step S28).

[4. Example of Flow of Processing Executed by Learning Device]
Next, an example of the flow of the learning process performed by the learning device 10 will be described using FIG. 7. FIG. 7 is a flowchart illustrating an example of the flow of learning processing according to the embodiment. First, the learning device 10 acquires a dialogue sentence (step S101), and extracts an utterance and a response from the dialogue sentence (step S102). Then, the learning device 10 performs pre-training on the learning device and the restoring device included in the first RNN so as to reproduce the response from the speech (step S103).

  Subsequently, the learning device 10 generates a second RNN using the pre-trained learning device and the new restoring device (step S104), and generates information indicating a label from the text of the labeled text. The second RNN is learned (step S105). Then, the learning device 10 generates a third RNN using the restored second RNN and the learning device included in the pre-trained first RNN (step S106), and the generated third RNN for measurement RNN To execute the measurement process (step S107), and the process ends.

[5. Modified example]
In the above, an example of the learning process by the learning device 10 has been described. However, the embodiments are not limited to this. Hereinafter, variations of the learning process performed by the learning device 10 will be described.

[5-1. About RNN for measurement]
In the example described above, the learning device 10 specifies the label to be attached to the text to which the label is to be attached, using the generated measurement RNN. However, the embodiments are not limited to this. For example, the learning device 10 may generate a program for applying a label from a sentence using the generated measurement RNN, and may output the generated program. Such a program can be executed by an information processing apparatus other than the learning apparatus 10, whereby the label can be specified from the text using the measurement RNN generated by the learning apparatus 10. That is, the measurement RNN generated by the learning device 10 may be used as a program parameter of a program for identifying a label from a sentence.

5-2. Device configuration〕
In the example described above, the learning device 10 executes the learning process and the measurement process in the learning device 10. However, the embodiments are not limited to this. For example, the learning device 10 may perform only the learning process, and another device may perform the measurement process. Also, the learning device 10 may store the dialogue data database 31 and the learning data database 32 in an external storage server.

[5-3. Other]
Further, among the processes described in the above embodiment, all or part of the process described as being automatically performed may be manually performed, or the process described as being manually performed. All or part of them can be performed automatically by known methods. In addition to the above, the processing procedures, specific names, and information including various data and parameters shown in the above text and drawings can be arbitrarily changed unless otherwise specified. For example, the various information shown in each figure is not limited to the illustrated information.

  Further, each component of each device illustrated is functionally conceptual, and does not necessarily have to be physically configured as illustrated. That is, the specific form of the distribution and integration of each device is not limited to the illustrated one, and all or a part thereof may be functionally or physically dispersed in any unit depending on various loads, usage conditions, etc. It can be integrated and configured. For example, the first learning unit 42 and the second learning unit 43 illustrated in FIG. 2 may be integrated.

  Moreover, it is possible to combine suitably each embodiment mentioned above in the range which does not contradict process content.

[6. program〕
Further, the learning device 10 according to the embodiment described above is realized by, for example, a computer 1000 configured as shown in FIG. FIG. 8 is a diagram illustrating an example of the hardware configuration. The computer 1000 is connected to an output device 1010 and an input device 1020, and a form in which an arithmetic device 1030, a primary storage device 1040, a secondary storage device 1050, an output IF (Interface) 1060, an input IF 1070, and a network IF 1080 are connected by a bus 1090. Have.

  The arithmetic unit 1030 operates based on a program stored in the primary storage unit 1040 or the secondary storage unit 1050, a program read from the input unit 1020, etc., and executes various processing. The primary storage device 1040 is a memory device, such as a RAM, temporarily storing data used by the computing device 1030 for various operations. The secondary storage device 1050 is a storage device in which data used by the arithmetic device 1030 for various calculations and various databases are registered, and is realized by a ROM (Read Only Memory), an HDD, a flash memory or the like.

  The output IF 1060 is an interface for transmitting information to be output to an output device 1010 that outputs various types of information such as a monitor and a printer. For example, USB (Universal Serial Bus), DVI (Digital Visual Interface), or the like It is realized by a connector of a standard such as HDMI (High Definition Multimedia Interface). The input IF 1070 is an interface for receiving information from various input devices 1020 such as a mouse, a keyboard, and a scanner, and is realized by, for example, USB or the like.

  The input device 1020 may be, for example, an optical recording medium such as a compact disc (CD), a digital versatile disc (DVD), a phase change rewritable disc (PD), a magneto-optical recording medium such as a magneto-optical disc (MO), or a tape. It may be a device that reads information from a medium, a magnetic recording medium, a semiconductor memory, or the like. The input device 1020 may also be an external storage medium such as a USB memory.

  The network IF 1080 receives data from another device via the network N and sends it to the computing device 1030, and transmits data generated by the computing device 1030 via the network N to the other device.

  The arithmetic device 1030 controls the output device 1010 and the input device 1020 via the output IF 1060 and the input IF 1070. For example, the processing unit 1030 loads a program from the input unit 1020 or the secondary storage unit 1050 onto the primary storage unit 1040 and executes the loaded program.

  For example, when the computer 1000 functions as the learning device 10, the computing device 1030 of the computer 1000 implements the function of the control unit 40 by executing the program loaded on the primary storage device 1040.

[7. effect〕
As described above, the learning device 10 causes the learning device and the restoring device of the first RNN to learn the relationship between the utterance and the response in the predetermined dialogue. Then, the learning device 10 generates a second RNN using at least one of the learning device or the restoring device in which the learning has been performed, and the relationship between the sentence and the label attached to the sentence is the second RNN. Let them learn. As described above, since the learning device 10 pre-trains the RNN using the hidden commonality included in the utterance and the response to the utterance, the accuracy of the label given to the sentence can be improved. In addition, when the learning data is abundant to some extent, it is possible to obtain an RNN having better classification accuracy (performance) than an RNN learned by the conventional learning method.

  In addition, when the speech is input to the learning device, the learning device 10 performs learning of the learning device and the restoring device such that the restoration device outputs a response to the speech. For example, the learning device 10 causes the learning device to learn the features possessed by each word along with the order in which the words included in the utterance appear in the utterance, and each word included in the response is extracted from the features encoded by the learning device. Train the decompressor to recover with the order in which the words appear in the response.

  Further, when the response is input to the learning device, the learning device 10 performs learning of the learning device and the recovery device such that the recovery device outputs an utterance corresponding to the response. For example, the learning device 10 causes the learning device to learn the features possessed by each word along with the order in which the words included in the response appear in the response, and each word included in the utterance is extracted from the features encoded by the learning device. Train the decompressor to recover with the order in which the words appear in the utterance.

  As a result of the above-described processing, the learning device 10 can perform pre-training of the RNN using the hidden commonality included in the utterance and the response to the utterance.

  Also, the learning device 10 generates a second RNN using a learner included in the first RNN subjected to the pre-training and a new restorer. Also, the learning device 10 generates a third RNN including a learning device included in the first RNN subjected to the pre-training and a restoration device included in the second RNN subjected to the learning using the learning data. Then, when the sentence to which the label is to be attached is received, the learning device 10 specifies the label to be attached to the received sentence using the third RNN. For this reason, the learning device 10 can prevent the deterioration of the accuracy of the label given to the text.

  Also, the learning device 10 generates a third RNN including a learning device included in the first RNN subjected to the pre-training and a restoration device included in the second RNN subjected to the learning using the learning data. Then, the learning device 10 causes the generated third RNN to learn the relationship between the text and the label attached to the text, and the label attached to the received text using the third RNN subjected to the learning. Identify For this reason, the learning device 10 can prevent the deterioration of the accuracy of the label given to the text.

  When the learning device 10 receives a sentence to which a label is to be attached, the learning device 10 uses the second RNN to specify a label to be attached to the received sentence. For this reason, the learning device 10 can prevent the deterioration of the accuracy of the label given to the text.

  Although some of the embodiments of the present application have been described above in detail based on the drawings, these are only examples, and various modifications can be made based on the knowledge of those skilled in the art, including the aspects described in the section of the description of the invention. It is possible to implement the invention in other improved forms.

  Also, the "section (module, unit)" described above can be read as "means" or "circuit". For example, the generation unit can be read as a generation unit or a generation circuit.

Reference Signs List 20 communication unit 30 storage unit 31 dialogue data database 32 learning data database 40 control unit 41 acquisition unit 42 first learning unit 43 second learning unit 44 generation unit 45 measurement unit 100 terminal device 200 input device

Claims (11)

A first learning unit that causes a learning device and a restoring device of the first recurrent neural network to learn the relationship between an utterance and a response in a predetermined dialogue;
A first generation unit that generates a second recurrent neural network using at least one of a learning device or a restoring device that has been learned by the first learning device;
And a second learning unit that causes the second recurrent neural network to learn the relationship between a sentence and a label given to the sentence. The first learning unit is characterized in that when the speech is input to the learning device, the first learning unit performs learning of the learning device and the restoration device such that the restoration device outputs a response to the speech. The learning device according to Item 1. The first learning unit causes the learning device to learn the features of each word along with the order in which the words included in the utterance appear in the utterance, and the features encoded by the learning device are included in the response. The learning device according to claim 2, wherein the decompressor is trained to recover each word in the response along with the order in which each word appears in the response. The first learning unit performs learning of the learning device and the restoring device such that when the response is input to the learning device, the restoring device outputs an utterance corresponding to the response. The learning device according to claim 1. The first learning unit causes the learning device to learn the features of each word along with the order in which the words included in the response appear in the response, and the features encoded by the learning device are included in the utterance. 5. The learning apparatus according to claim 4, wherein the decompressor is trained so as to recover each word that is generated along with the order in which each word appears in the utterance. The first generation unit generates the second recurrent neural network using a learning device that has been learned by the first learning unit and a new restoration device. The learning device according to any one. A second generation unit that generates a third recurrent neural network including a learning unit learned by the first learning unit and a restoration unit included in the second recurrent neural network learned by the second learning unit;
When a sentence to which a label is to be attached is received, the third recurrent neural network generated by the second generation unit is used to specify a label to be attached to the received sentence. The learning device according to claim 6, characterized in that A second generation unit that generates a third recurrent neural network including a learning unit learned by the first learning unit and a restoration unit included in the second recurrent neural network learned by the second learning unit;
A third learning unit that causes the third recurrent neural network generated by the second generation unit to learn the relationship between the sentence and the label attached to the sentence;
When a sentence to which a label is to be attached is received, the identification unit for identifying a label to be attached to the accepted sentence using the third recurrent neural network learned by the third learning unit The learning apparatus according to claim 6, comprising: When a sentence to which a label is to be attached is received, the identification unit for identifying a label to be attached to the accepted sentence using the second recurrent neural network learned by the second learning unit The learning apparatus according to claim 6, comprising: A learning method executed by the learning device,
A first learning step of causing the learning device and the restoring device of the first recurrent neural network to learn the relationship between an utterance and a response in a predetermined dialogue;
A first generation step of generating a second recurrent neural network using at least one of a learner and a restorer trained in the first learn step;
And a second learning step of causing the second recurrent neural network to learn the relationship between a sentence and a label attached to the sentence. A first learning procedure for causing a computer to learn the relationship between an utterance and a response in a predetermined dialogue on a learning device and a restoring device included in the first recurrent neural network;
A first generation procedure for generating a second recurrent neural network using at least one of a learner and a restorer trained by the first learn procedure;
A second learning procedure for causing the second recurrent neural network to learn the relationship between a sentence and a label given to the sentence.

Images