Intention analysis method and system

Abstract

The present invention discloses an intent analysis method and system, wherein a part-of-speech tagging model is obtained by training with domain words, and new related words can be updated to a reference word database and associated with related domain words, so that the part-of-speech tagging model can quickly adapt to new vocabulary and respond to service needs quickly; part-of-speech tagging models for several fields can also be obtained by training with domain words in different fields, and domain expression sentences and reference expression sentences in different fields are compared respectively, and a reference expression sentence with the highest matching degree is obtained as an intent determination result, thereby effectively avoiding misrecognition of intent caused by ambiguous words and improving the accuracy of intent determination.

Description

Technical Field

[0001] The present invention belongs to the technical field of intention recognition, and in particular, relates to an intention analysis method and system. Background Art

[0002] Artificial intelligence is gradually penetrating into people's daily lives, and natural language processing plays a pivotal role in this. How to quickly understand the user's intentions and provide corresponding product services is crucial to the development of enterprises.

[0003] At present, in order to improve the efficiency of understanding user intentions, a large number of large and complex models need to be continuously trained. Faced with the diverse vocabulary in life, the model needs to re-learn new vocabulary, which leads to long model training time and high maintenance cost. At the same time, ambiguous words are inevitable in the user's language expression, which easily leads to intent recognition errors. Summary of the invention

[0004] In order to overcome the deficiencies of the prior art, the purpose of the present invention is to provide an intent analysis method and system to solve the problem that the current intent recognition model cannot quickly adapt to new vocabulary and is prone to misrecognition of ambiguous words.

[0005] The purpose of the present invention is achieved by the following technical solutions:

[0006] An intention analysis method, comprising:

[0007] Establishing a reference word and sentence database, the reference word and sentence database includes domain words, related words associated with the domain words, and reference expression sentences with intention marks;

[0008] According to the domain words, a part-of-speech tagging model of the domain to which the words belong is obtained through training;

[0009] Acquire language information input by a user, segment the language information, and respectively annotate the domain words in the language information based on a plurality of part-of-speech tagging models to obtain a plurality of domain expression sentence patterns;

[0010] Each of the domain expression sentences is compared with the reference expression sentence in the same domain, the reference expression sentence with the highest matching degree is used as the intention determination result, and the intention tag is obtained based on the reference expression sentence of the intention determination result.

[0011] Furthermore, the step of establishing a reference word and sentence database includes:

[0012] The domain word is obtained, and the related words associated with the domain word are obtained according to the domain word.

[0013] Furthermore, the step of establishing a reference word and sentence database further includes:

[0014] Obtain corresponding domain expression sentence patterns in several fields;

[0015] The related words appearing in the domain expression sentence pattern are replaced with the domain words associated with the related words to obtain the reference expression sentence pattern.

[0016] Furthermore, after the step of replacing the related words appearing in the domain expression sentence pattern with the domain words associated with the related words to obtain the reference expression sentence pattern, the method further includes:

[0017] The intention tag is added to the reference expression sentence pattern according to the service that the reference expression sentence pattern needs to provide.

[0018] Furthermore, the step of obtaining a part-of-speech tagging model of the domain to which the domain word belongs based on training also includes:

[0019] A basic corpus is generated according to the domain words and the domain expression sentence patterns, and the part-of-speech tagging model is obtained by training with the basic corpus.

[0020] Furthermore, the step of tagging the domain words in the language information based on the part-of-speech tagging model further includes:

[0021] The part-of-speech tagging model identifies relevant words in the field to which it belongs and tags them as corresponding words in the field.

[0022] Furthermore, the step of tagging the domain words in the language information based on the part-of-speech tagging model further includes:

[0023] The new related words are associated with the domain words related thereto, and the part-of-speech tagging model identifies the new related words appearing in the language information and tags them as the domain words.

[0024] Furthermore, after selecting the reference expression sentence pattern that is in the same field as the field expression sentence pattern and has the highest matching degree, and obtaining the intention marking step based on the reference expression sentence pattern, the method further includes:

[0025] Relevant words in the domain expression sentence are extracted, and the relevant words and intent tags are sent to the server for processing.

[0026] In order to solve the same technical problem, the present invention also provides an intention analysis system, comprising:

[0027] A reference word and sentence database, used to store reference expression sentences with intention tags;

[0028] A part-of-speech tagging model generation module is used to obtain a part-of-speech tagging model based on domain word training;

[0029] A user input module, used to obtain language information input by the user;

[0030] A word segmentation module, used for performing word segmentation on the language information;

[0031] A domain expression sentence generation module, used for annotating the domain words in the language information based on the part-of-speech tagging model to obtain a domain expression sentence;

[0032] The comparison module is used to screen out the reference expression sentence pattern that is in the same field as the field expression sentence pattern and has the highest matching degree, and obtain the intention tag based on the reference expression sentence pattern.

[0033] Furthermore, the reference phrase database includes:

[0034] A domain word storage unit, used to store the domain words;

[0035] A related word storage unit, used for storing the related words and associating the related words with the domain words related thereto;

[0036] A domain expression sentence pattern storage unit, used for storing domain expression sentence patterns;

[0037] The reference expression sentence pattern generating unit is used to replace the related words appearing in the domain expression sentence pattern with the domain words associated with the related words to obtain the reference expression sentence pattern.

[0038] Compared with the prior art, the present invention has the following beneficial effects:

[0039] (1) The part-of-speech tagging model is trained based on domain words, and can update new related words to the reference word database and associate them with related domain words, so that the part-of-speech tagging model can quickly adapt to new words and respond to service needs quickly;

[0040] (2) Part-of-speech tagging models for several fields are obtained based on domain words from different fields. Domain expression sentences and reference expression sentences from different fields are compared respectively, and the reference expression sentence with the highest matching degree is obtained as the intention judgment result, thereby effectively avoiding the misrecognition of intention caused by ambiguous words and improving the accuracy of intention judgment. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 A flow chart of the method for analyzing the intent of the invention;

[0042] Figure 2 A schematic diagram of the structure of the system for analyzing the intent of the invention.

[0043] In the figure:

[0044] 100. Reference word and sentence database; 110. Domain word storage unit; 120. Related word storage unit; 130. Domain expression sentence pattern storage unit; 140. Reference expression sentence pattern generation unit; 200. Part-of-speech tagging model generation module; 300. User input module; 400. Word segmentation module; 500. Domain expression sentence pattern generation module; 600. Comparison module. DETAILED DESCRIPTION

[0045] In order to facilitate the understanding of the present invention, the present invention will be described more fully below with reference to the relevant drawings. Preferred embodiments of the present invention are provided in the drawings. However, the present invention can be implemented in many different forms and is not limited to the embodiments described herein. On the contrary, the purpose of providing these embodiments is to make the understanding of the disclosure of the present invention more thorough and comprehensive.

[0046] It should be noted that when an element is referred to as being "fixed to" another element, it may be directly on the other element or there may be a central element. When an element is considered to be "connected to" another element, it may be directly connected to the other element or there may be a central element at the same time. The terms "vertical", "horizontal", "left", "right" and similar expressions used herein are for illustrative purposes only.

[0047] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art of the present invention. The terms used herein in the specification of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention. The term "and / or" used herein includes any and all combinations of one or more related listed items.

[0048] Combination Figure 1 As shown, the process of the intention analysis method of the present invention is schematically shown, including:

[0049] S100, establishing a reference word and sentence database; specifically,

[0050] S101. Obtain domain words, and obtain related words associated with the domain words

[0051] For example, in the field of cooking, the following domain words can be obtained: dish name / ndish, ingredients / nfood, people / npeople, taste / ataste, function / nfunction, festival / tfestival, cuisine / ncuisine, etc. Among them, the "dish name / ndish" in "dish name / ndish" is the domain word, and "ndish" is the annotation word of the domain word. Similarly, the "ingredients / nfood" in "ingredients / nfood" is the domain word, and "nfood" is the annotation word of the domain word.

[0052] The following related words can be obtained based on the above-mentioned domain words in the cooking field:

[0053] Dish name: glutinous rice chicken, beer duck, hot and sour potato shreds, etc.

[0054] Ingredients / nfood: beef, potatoes, tomatoes, etc.;

[0055] People: old people, children, pregnant women, etc.

[0056] Taste / ataste: sour, sweet, bitter, spicy, etc.;

[0057] Function: nourishing the skin and hair, darkening hair and improving eyesight, preventing and treating edema, etc.

[0058] Festival: Lantern Festival, Dragon Boat Festival, Chinese Valentine's Day, etc.

[0059] Cuisine: Cantonese, Guangdong, Teochew, etc.

[0060] For another example, in the field of music, the following domain words can be obtained: song / nsong, singer / nsinger, language / nlang, genre / ngenre, etc.; among them, the "song" in "song / nsong" is the domain word, and "nsong" is the tag word of the domain word. Similarly, the "singer" in "singer / nsinger" is the domain word, and "nsinger" is the tag word of the domain word.

[0061] The following related words can be obtained based on the domain words in the music field:

[0062] Song / nsong: song name 1, song name 2, song name 3, etc.;

[0063] Singer / nsinger: singer name 1, singer name 2, singer name 3, etc.;

[0064] Language / nlang: Chinese, Cantonese, English, etc.;

[0065] Genre: Folk, Rock, Pop, etc.

[0066] Of course, other fields can refer to the acquisition of domain words and related words in the above-mentioned cooking field and music field, which will not be repeated here.

[0067] S102, obtaining corresponding domain expression sentence patterns in a number of domains, where the domain expression sentence patterns may be commonly used expression sentence patterns in the domains to which they belong;

[0068] For example, in the field of cooking, the following expression sentences can be obtained:

[0069] How to make spicy and sour potato shreds delicious / what are the simple methods of making beer duck, etc.;

[0070] What dishes can be made with beef / what non-spicy dishes can be made with potatoes and tomatoes, etc.;

[0071] What dishes are suitable for the elderly / what can pregnant women eat, etc.;

[0072] What are the ingredients for dark hair / what are the good foods for thickening and nourishing hair, etc.

[0073] What dishes are recommended for Chinese Valentine's Day / What do people eat on Lantern Festival?

[0074] What are the Cantonese dishes / Chaozhou dishes, etc.

[0075] For example, in the field of music, the following expression patterns can be obtained:

[0076] I want to listen to a song title 1 / sing a song title 2, etc.;

[0077] Play a song by singer 1 / play a song by singer 2, etc.

[0078] I want to listen to Chinese songs / I want to listen to English songs sung by singer name 2, etc.;

[0079] Let's have a pop song / I want to hear a ballad, etc.

[0080] For other fields, the acquisition of domain expression sentence patterns in the above-mentioned cooking and music fields can be referred to, and will not be repeated here.

[0081] S103, replacing the related words appearing in the domain expression sentence pattern with domain words associated with the related words to obtain a reference expression sentence pattern;

[0082] For example, in the field of cooking, the field expression sentence pattern is processed to obtain the reference expression sentence pattern:

[0083] How to make {dish name} delicious / What are the simple ways to make {dish name}, etc.;

[0084] What dishes can be made with {ingredients} / What non-spicy dishes can be made with {ingredients} and {ingredients}, etc.;

[0085] What dishes are suitable for {people} to eat / what {people} can eat, etc.;

[0086] What are the ingredients with {effects} / what are the good things to eat with {effects}, etc.;

[0087] What dishes are recommended for {festival} / What do people eat at {festival}, etc.;

[0088] What {cuisines} are there / What {cuisines} are there, etc.

[0089] Another example: In the field of music, the domain expression sentence pattern is processed to obtain the reference expression sentence pattern:

[0090] I want to listen to {song} / sing a {song}, etc.;

[0091] Play {singer}'s song / Play a {singer}'s song, etc.;

[0092] I want to listen to songs in {language} / I want to listen to songs in {language} sung by {singer}, etc.;

[0093] Listen to a {genre} song / I'd like to listen to {genre}, etc.

[0094] For other fields, the processing methods of the field expression sentences in the above-mentioned cooking and music fields can be referred to, and no further details will be given here.

[0095] S104, adding an intention tag to the reference expression sentence according to the service required to be provided by the reference expression sentence;

[0096] For example, in the cooking domain, add intent markers to the following reference expressions:

[0097] How to make {dish name} delicious / What are the simple ways to make {dish name}, etc.;

[0098] What dishes can be made with {ingredients} / What non-spicy dishes can be made with {ingredients} and {ingredients}, etc.;

[0099] What dishes are suitable for {people} to eat / what {people} can eat, etc.;

[0100] What dishes are recommended for {festival} / What do people eat at {festival}, etc.;

[0101] What {cuisines} are there / What {cuisines} are there, etc.;

[0102] The intention of the reference expression sentence above is marked as: "ask what the name of the dish is / what-dish".

[0103] What are the ingredients with {effects} / what are the good things to eat with {effects}, etc.;

[0104] The intention of the reference expression sentence above is marked as: "asking what ingredients / what-food".

[0105] Another example: In the field of music, add intent tags to the following reference expressions:

[0106] I want to listen to {song} / sing a {song}, etc.;

[0107] Play {singer}'s song / Play a {singer}'s song, etc.;

[0108] I want to listen to songs in {language} / I want to listen to songs in {language} sung by {singer}, etc.;

[0109] Play a {genre} song / I want to listen to {genre}, etc.;

[0110] The intention of the reference expression sentence above is marked as: "play song / what-song".

[0111] For other fields, the intention tag adding operation in the above-mentioned cooking field and music field can be referred to, and will not be repeated here.

[0112] S200, obtain a part-of-speech tagging model for the domain to which it belongs based on domain word training; specifically,

[0113] Generate basic corpus based on domain words and domain expression sentence patterns, and use the basic corpus to train a part-of-speech tagging model. For the training of the part-of-speech tagging model, the existing part-of-speech tagging model training method in the field of deep learning can be used. The part-of-speech tagging effect of the part-of-speech tagging model is as follows:

[0114] After part-of-speech tagging, the result of "What are the methods of making pickled fish" is: "pickled fish_ndish, have_v, which_r, method_n";

[0115] After part-of-speech tagging, the result of “What non-spicy dishes can be made with potatoes and beef” is: “potatoes_nfood, and_c, beef_nfood, can_v, make_v, what_r, not_d, spicy_ataste, of_u, dish_n”;

[0116] After part-of-speech tagging, “I want to listen to song 1 by singer 1” is obtained as follows: “I_r, want_v, listen_v, singer 1_nsinger, of_u, song 1_nsong”.

[0117] Among them, the part-of-speech tagging model in a certain field has different effects on the part-of-speech tagging in different fields. For example, the part-of-speech tagging model in the cooking field has different effects on the following two sentence types:

[0118] After part-of-speech tagging, the result of "What are the methods of making pickled fish" is: "pickled fish_ndish, have_v, which_r, method_n";

[0119] “I want to listen to song 1 by singer 1” is processed by part-of-speech tagging to get: “I_r, want_v, listen_v, singer_nr,_u, song_n”.

[0120] The part-of-speech tagging model in the music field has the following two sentence types:

[0121] "I want to listen to the song name 1 of singer name 1" is processed by part-of-speech tagging to get: "I_r, want_v, listen_v, singer name 1_nsinger, of_u, song name 1_nsong";

[0122] After part-of-speech tagging, the sentence “What are the methods of cooking pickled fish” is obtained as follows: “pickled fish_n, have_v, which_r, methods_n”.

[0123] It can be seen that the part-of-speech tagging model in the cooking field can recognize the relevant word "sour fish" in the cooking field and mark it as the domain word "dish name / ndish", but cannot recognize the relevant words "singer name 1" and "song name 1" in the music field; similarly, the part-of-speech tagging model in the music field can recognize the relevant word "singer name 1" in the music field and mark it as the domain word "singer / nsinger" and the related word "song name 1" and mark it as the domain word "song / nsong". This is because each part-of-speech tagging model is trained according to the domain words and domain expression sentence patterns of its own field.

[0124] S300, obtaining language information input by a user, segmenting the language information, and annotating domain words in the language information based on a plurality of part-of-speech tagging models to obtain a plurality of domain expression sentences; specifically,

[0125] The language information input by the user can be collected through input devices such as microphones and keyboards;

[0126] If the language information input by the user is voice, the user's voice is first converted into recognized text, and then the recognized text is segmented;

[0127] The part-of-speech tagging model identifies the relevant words in the field to which it belongs and tags them as corresponding field words to obtain the field expression sentence pattern; for example:

[0128] In the cooking domain, the user inputs language information "What are the methods of cooking pickled fish?", and the part-of-speech tagging model recognizes and tags it to get "pickled fish_ndish, have_v, which_r, method_n". Combined with the domain word "dish name / ndish", the domain expression sentence "What are the methods of cooking {dish name}" can be obtained;

[0129] In the field of music, the user inputs the language information "I want to listen to AA and BB's C1's D1", and after recognition and annotation by the part-of-speech tagging model, we get "I_r, want_v, listen_v,AA and BB_nsinger,'s_u,C1's D1_nsong". Combined with the domain words "song / nsong" and "singer / nsinger", we get the domain expression sentence "I want to listen to {singer}'s {song}".

[0130] Of course, with the development and progress of society, there will be more and more related words associated with domain words. For example, in the field of music, the part-of-speech tagging model has not yet associated the new related word "C1's D1" with the domain word "song / nsong", so the part-of-speech tagging model in the music field will annotate the language information "I want to listen to AA and BB's C1's D1" as "I_r, want_v, listen_v, AA and BB_nsinger, of_u, C1_n, of_u, D1_n". In order to quickly adapt to the development of society, new related words can be associated with related domain words. The part-of-speech tagging model identifies new related words appearing in the language information and annotates them as related domain words. For example, after the new related word "C1's D1" is associated with the domain word "song / nsong", the part-of-speech tagging model in the music field will annotate the language information "I want to listen to AA and BB's C1's D1" as "I_r, want_v, listen_v, AA and BB_nsinger, of_u, C1's D1_nsong".

[0131] S400, comparing each domain expression sentence with a reference expression sentence in the same domain, taking the reference expression sentence with the highest matching degree as the intention determination result, and obtaining an intention tag based on the reference expression sentence of the intention determination result;

[0132] Specifically, after the user inputs language information, the language information is identified and annotated using part-of-speech tagging models in several fields to obtain several field expression sentences, and each field expression sentence is compared with a reference expression sentence in the same field to detect the matching degree. If a field expression sentence has the highest matching degree with the reference expression sentence, it is used as the intention determination result, and the reference expression sentence based on the intention determination result can obtain the corresponding intention tag;

[0133] For example:

[0134] The user inputs the language information "I want to listen to AA and BB's C1 and D1";

[0135] The part-of-speech tagging model in the music field identifies and tags the language information to obtain "I_r, want_v, listen_v, AA and BB_nsinger, of_u, C1 ofD1_nsong", and then obtains the domain expression sentence "I want to listen to {singer}'s {song}";

[0136] The part-of-speech tagging model in the cooking field identifies and tags the language information to obtain "I_r, want_v, listen_v, AA_n, and_u, BB_n, of_u, C1_n, of_u, D1_n", and then obtains the domain expression sentence "I want {noun} and {noun} of {noun} of {noun}";

[0137] The above-mentioned domain expression sentence "I want to listen to {song} by {singer}" is compared with several reference expression sentences in the same music field, and it is found that the domain expression sentence has the highest matching degree with the reference expression sentence "play {song} by {singer}" in the music field, with a matching similarity of 90%;

[0138] The above-mentioned domain expression sentence "I think {noun} and {noun}'s {noun}'s {noun}" is compared with the reference expression sentence belonging to the cooking field, and it is found that the domain expression sentence and the reference expression sentence in the cooking field "How to make {dish name} delicious" have the highest matching degree, and the matching similarity is 10%; among which, the above-mentioned sentence matching degree calculation method can be calculated using the currently existing sentence editing distance or the sentence matching model publicly available in the deep learning field.

[0139] Obviously, the matching similarity between the domain expression sentence "I want to listen to {singer}'s {song}" and the reference expression sentence "play {singer}'s {song}" in the music domain is higher than the matching similarity between the domain expression sentence "I want {noun} and {noun}'s {noun}'s {noun}" and the reference expression sentence "How to cook {dish name} to make it delicious" in the cooking domain. It can be determined that the language information "I want to listen to AA and BB's C1's D1" belongs to the music domain, and the language information should refer to the intention tag of the reference expression sentence "play {singer}'s {song}", that is, the intention tag is "play song / what-song".

[0140] S500, extract relevant words in the domain expression sentence, and send the relevant words and intent tags to the server for processing. In the above S400 example, the relevant words to be extracted are "AA and BB", "D1 of C1", and the intent tag is "play song / what-song". The server can find the song D1 of C1 by singers AA and BB based on the above relevant words and intent tags, and play the song.

[0141] In order to solve the same technical problem, the present invention also provides an intention analysis system, such as Figure 2 ,include:

[0142] A reference word and sentence database 100, for storing reference expression sentences with intention tags;

[0143] The reference phrase database 100 includes:

[0144] A domain word storage unit 110, used to store domain words;

[0145] A related word storage unit 120, used to store the related words and associate the related words with related domain words;

[0146] A domain expression sentence pattern storage unit 130, used to store domain expression sentence patterns;

[0147] A reference expression sentence pattern generating unit 140 is used to replace the related words appearing in the domain expression sentence pattern with domain words associated with the related words to obtain a reference expression sentence pattern;

[0148] A part-of-speech tagging model generation module 200 is used to obtain a part-of-speech tagging model based on domain word training;

[0149] The user input module 300 is used to obtain language information input by the user;

[0150] A word segmentation module 400 is used to segment language information;

[0151] A domain expression sentence generation module 500 is used to annotate domain words in language information based on a part-of-speech tagging model to obtain a domain expression sentence;

[0152] The comparison module 600 is used to screen out a reference expression sentence pattern that is in the same domain as the domain expression sentence pattern and has the highest matching degree, and obtain an intention tag based on the reference expression sentence pattern.

[0153] Compared with the prior art, the present invention has the following beneficial effects:

[0154] (1) The part-of-speech tagging model is trained based on domain words, and can update new related words to the reference word database and associate them with related domain words, so that the part-of-speech tagging model can quickly adapt to new words and respond to service needs quickly;

[0155] (2) Part-of-speech tagging models for several fields are obtained based on domain words from different fields. Domain expression sentences and reference expression sentences from different fields are compared respectively, and the reference expression sentence with the highest matching degree is obtained as the intention judgment result, thereby effectively avoiding the misrecognition of intention caused by ambiguous words and improving the accuracy of intention judgment.

[0156] The above descriptions are merely embodiments of the present invention and are not intended to limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made using the contents of the present invention specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. An intention analysis method, characterized in that: include: Establishing a reference word and sentence database, the reference word and sentence database includes domain words, related words associated with the domain words, and reference expression sentences with intention marks; According to the domain words, a part-of-speech tagging model of the domain to which the words belong is obtained through training; Acquire language information input by a user, segment the language information, and respectively annotate the domain words in the language information based on the part-of-speech tagging models of several domains to obtain several domain expression sentence patterns; Compare each of the domain expression sentences with the reference expression sentence in the same domain, take the reference expression sentence with the highest matching degree as the intention determination result, and obtain the intention tag based on the reference expression sentence of the intention determination result; The step of comparing each of the domain expression sentences with the reference expression sentences in the same domain and taking the reference expression sentence with the highest matching degree as the intention determination result comprises: Each domain expression sentence is compared with the reference expression sentence in the same domain to detect the degree of match. If a domain expression sentence has the highest degree of match with the reference expression sentence, it is taken as the intention judgment result.

2. The intention analysis method according to claim 1, characterized in that: The step of establishing a reference word database comprises: The domain word is obtained, and the related words associated with the domain word are obtained according to the domain word.

3. The intention analysis method according to claim 2, characterized in that: The step of establishing a reference phrase database also includes: Obtain corresponding domain expression sentence patterns in several fields; The related words appearing in the domain expression sentence pattern are replaced with the domain words associated with the related words to obtain the reference expression sentence pattern.

4. The intention analysis method according to claim 3, characterized in that: After the step of replacing the related words appearing in the domain expression sentence pattern with the domain words associated with the related words to obtain the reference expression sentence pattern, the method further includes: The intention tag is added to the reference expression sentence pattern according to the service that the reference expression sentence pattern needs to provide.

5. The intention analysis method according to claim 3, characterized in that: The step of obtaining a part-of-speech tagging model for the domain to which the domain word belongs based on training also includes: A basic corpus is generated according to the domain words and the domain expression sentence patterns, and the part-of-speech tagging model is obtained by training with the basic corpus.

6. The intention analysis method according to claim 1, characterized in that: The step of tagging the domain words in the language information based on the part-of-speech tagging model further includes: The part-of-speech tagging model identifies relevant words in the field to which it belongs and tags them as corresponding words in the field.

7. The intention analysis method according to claim 1, characterized in that: The step of tagging the domain words in the language information based on the part-of-speech tagging model further includes: The new related words are associated with the domain words related thereto, and the part-of-speech tagging model identifies the new related words appearing in the language information and tags them as the domain words.

8. The intention analysis method according to claim 1, characterized in that: After selecting the reference expression sentence pattern that is in the same field as the field expression sentence pattern and has the highest matching degree, and obtaining the intention tagging step based on the reference expression sentence pattern, the method further includes: Relevant words in the domain expression sentence are extracted, and the relevant words and intent tags are sent to the server for processing.

9. An intention analysis system, characterized in that: include: A reference word and sentence database, used to store reference expression sentences with intention tags; The reference word and sentence database includes domain words, related words associated with the domain words, and reference expression sentences with intention marks; A part-of-speech tagging model generation module is used to obtain a part-of-speech tagging model based on domain word training; A user input module, used to obtain language information input by the user; A word segmentation module, used for performing word segmentation on the language information; A domain expression sentence generation module, used for annotating the domain words in the language information based on the part-of-speech tagging models of several domains to obtain several domain expression sentence patterns; A comparison module, used for comparing each of the domain expression sentences with the reference expression sentence in the same domain, taking the reference expression sentence with the highest matching degree as the intention determination result, and obtaining the intention tag based on the reference expression sentence of the intention determination result; The comparison module is also used to compare each domain expression sentence with a reference expression sentence in the same domain to detect the matching degree. If a domain expression sentence has the highest matching degree with the reference expression sentence, it is used as the intention determination result.

10. The intention analysis system according to claim 9, characterized in that: The reference phrase database includes: A domain word storage unit, used to store the domain words; A related word storage unit, used for storing the related words and associating the related words with the domain words related thereto; A domain expression sentence pattern storage unit, used for storing domain expression sentence patterns; The reference expression sentence pattern generating unit is used to replace the related words appearing in the domain expression sentence pattern with the domain words associated with the related words to obtain the reference expression sentence pattern.

Images