A method for improving information exchange transmission efficiency in a robot conversation

Abstract

The application discloses a method for improving information exchange transmission efficiency in a robot conversation, and belongs to the field of robot interactive information processing. The method comprises the following steps: constructing a word abbreviation rule and an equivalent replacement dictionary table; identifying a commonly known object; and expressing the commonly known object by using a reference and an abbreviation word in a robot conversation process based on the word abbreviation rule and the equivalent replacement dictionary table, so that the length of the conversation text is reduced. The application adopts a natural language processing technology, discovers consensus knowledge by mining information interactive historical data between robots, and can realize reduction of information interactive transmission data amount by means of the consensus knowledge, and further reduce the bandwidth demand of a communication transmission line.

Description

Technical Field

[0001] This invention relates to the field of robot interactive information processing, and in particular to a method for improving the efficiency of information exchange and transmission in robot dialogue. Background Technology

[0002] Robots have been widely used in fields such as intelligent transportation, video surveillance, and environmental monitoring. With the continuous advancement of intelligent technology, robots can not only detect and perceive surrounding objects and environmental parameters, but also share the processing results of detection and perception with other robots to achieve a unified understanding of global information.

[0003] Information sharing requires bandwidth, and reducing the amount of data transmitted is crucial when bandwidth resources are limited. As robots become more intelligent, they gain the ability to generate, interact with, and process information to form knowledge. Considering the limited bandwidth resources of communication networks, technical means are needed in the interaction process to reduce the amount of data exchanged between robots while ensuring the integrity of the information.

[0004] Traditional information compression methods are based on information entropy theory. First, a dictionary is created to record the correspondence between various text strings and their corresponding encoded characters. Frequently occurring long text strings are replaced with shorter encoded characters. Then, the dictionary is used to replace the original text, thus compressing the text length. However, once the dictionary is created, it must be shared among all parties exchanging information, making modification and replacement inconvenient. Furthermore, the encoded information is not easily readable. Summary of the Invention

[0005] To effectively reduce the bandwidth requirements for information transmission, this invention proposes a method to improve the efficiency of information exchange in robot dialogue. This method analyzes the historical text data of the dialogue between the two parties, identifies the background knowledge shared by both sides, and then replaces the original text with appropriate pronouns, thereby reducing the length of the original text.

[0006] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0007] A method for improving the efficiency of information exchange and transmission in robot dialogue, characterized by comprising the following steps:

[0008] Step 1: Construct a dictionary of word abbreviation rules and equivalent substitutions, specifically as follows:

[0009] (1.1) Automatically discover typical referential expression patterns in massive corpus texts by using language models;

[0010] (1.2) Each reference expression is parsed through manual analysis to form an entity reference equivalence set corresponding to the reference expression. Then, the various entity reference equivalence sets are analyzed to form word abbreviation rules and equivalent substitution dictionary table.

[0011] Step 2, identify commonly known objects, specifically as follows:

[0012] (2.1) Segment the dialogue history text of both parties and mark its parts of speech;

[0013] (2.2) Based on the word segmentation results, each word is taken as the starting word and concatenated with different numbers of subsequent adjacent words to form strings of various lengths;

[0014] (2.3) Delete strings that end with adverbs or particles;

[0015] (2.4) Count the frequency of each string and select the strings whose frequency exceeds the threshold;

[0016] (2.5) Use grammatical and semantic methods to identify locative words, names of people, names of organizations and factual descriptions as common knowledge objects; common knowledge objects are places, titles, names of objects, names of organizations and factual descriptions mentioned by both parties in the dialogue, and their forms are words, phrases, sentences and sentence groups;

[0017] Step 3: Based on word abbreviation rules and equivalent substitution dictionary, common knowledge objects are expressed using references and abbreviations during robot dialogue, thereby reducing the length of the dialogue text.

[0018] The beneficial effects of this invention are as follows:

[0019] 1. This invention uses natural language processing technology to discover consensus knowledge by mining historical data of information interaction between robots. With the help of consensus knowledge, the amount of data transmitted in information interaction is reduced, thereby reducing the bandwidth requirements of communication transmission lines.

[0020] 2. Based on natural language processing technology, this invention analyzes the historical data of information interaction between robots, identifies words, phrases, sentences and sentence groups commonly used by the parties involved in the information interaction, maps them to pronouns, and uses pronouns to achieve a concise expression of complex information, thereby compressing the text length.

[0021] 3. This invention draws upon the mechanism of efficient information exchange through the use of pronouns and abbreviations in human natural language dialogue, leveraging the context of the current conversation. Utilizing natural language processing technology, it analyzes historical data of information exchange between robots to identify commonly used words, phrases, sentences, and sentence groups as shared knowledge. These shared knowledge objects are then appropriately pronouncing and abbreviating to reduce the length of the dialogue text and lower communication bandwidth requirements. Furthermore, applying the mechanism of abbreviations used in human natural language dialogue to robot information exchange facilitates human understanding of the information exchanged between robots. Attached Figure Description

[0022] Figure 1 This is a flowchart of the present invention. Detailed Implementation

[0023] The present invention will now be further described with reference to the accompanying drawings.

[0024] A method for improving the efficiency of information exchange and transmission in robot dialogue includes the following steps:

[0025] 1. Construct a dictionary of word abbreviation rules and equivalent substitutions.

[0026] (1.1) Discover typical reference expression patterns in the text

[0027] Using language models, typical referential expression patterns appearing in massive corpora can be automatically discovered.

[0028] (1.2) Forming string abbreviation rules and equivalent substitution dictionary table

[0029] Each reference is parsed manually to form an entity reference equivalence set corresponding to the reference. Then, the various entity reference equivalence sets are analyzed to form word abbreviation rules and an equivalent substitution dictionary table.

[0030] 2. Identify common objects

[0031] (2.1) Segment the dialogue history text of both parties and mark the parts of speech.

[0032] (2.2) Based on the word segmentation results, each word is used as the starting word in turn, and different numbers of subsequent adjacent words are concatenated to form strings of various lengths.

[0033] (2.3) Delete strings that end with adverbs and particles.

[0034] (2.4) Count the frequency of each string and select the strings whose frequency exceeds the threshold.

[0035] (2.5) Identify the location words, personal names, organization names, and factual descriptions therein using syntactic and semantic methods as common knowledge objects. The common knowledge objects are the locations, appellations, object names, organization names, and factual descriptions mentioned by both parties in the conversation, and their forms are words, phrases, sentences, and sentence groups.

[0036] 3. Perform anaphora and abbreviation replacement

[0037] Based on the word abbreviation rules and the equivalent replacement dictionary table, during the robot conversation process, express the common knowledge objects using anaphora and abbreviated words, thereby reducing the length of the conversation text.

[0038] This method is mainly used to solve the problem of efficient information transfer between robots and is applicable to the information interaction scenarios between robots in various industries.

[0039] The following takes the example of intelligent monitoring probes in the urban road monitoring scenario interacting with each other's road condition perception results for a more detailed explanation:

[0040] A method for improving the efficiency of information exchange and transfer in robot conversations includes the following steps: [[ID= 17]]

[0041] 1. Construct word abbreviation rules and an equivalent replacement dictionary table

[0042] The purpose of this step is to find various anaphora and abbreviation usage examples in human natural language texts from a large amount of corpus texts. Through manual analysis, word abbreviation rules and an equivalent replacement dictionary table that can be used during robot interaction are formed. For this purpose, it is achieved through the following two steps.

[0043] (1.1) Discover typical anaphora expression patterns in the text

[0044] The more typical anaphora expression patterns discovered from a large amount of corpus texts, the more comprehensive the abbreviation rules and the equivalent replacement dictionary table formed in step (1.2) will be. For this purpose, this method uses a language model to automatically discover typical anaphora expression patterns that appear in a large amount of corpus texts.

[0045] There are mainly three typical forms of anaphora in Chinese:

[0046] Personal pronouns: Road monitoring found that [red car with license plate number Beijing C1234X] appeared near Beijing West Station at 10:20 am. One minute later, [it] appeared near Beijing South Station again.

[0047] Demonstrative pronouns: Road monitoring found that [the traffic flow density in the outbound direction of xxx Road leading to the Dukakan Expressway intersection is yyyyyy]. [This road condition] indicates that there will be traffic congestion at the Dukakan Expressway intersection.

[0048] Fact description: Road surveillance footage revealed that [water accumulation on a certain underpass caused slow traffic], which is the root cause of [traffic congestion on xxx Road]. This [root cause] has occurred 5 times this month.

[0049] The three typical referential phenomena mentioned above in the corpus text data were manually labeled, with labels including "personal referentiality," "demonstrative referentiality," and "factual referentiality." At least 1000 examples of each label were generated to form a training dataset. A BERT-BiLSTM-CRF classifier was trained using this dataset, and then used to label sentences, sentence groups, and paragraphs in the actual text data to identify the three typical referential patterns mentioned above in the actual text data.

[0050] (1.2) Forming string abbreviation rules and equivalent substitution dictionary table

[0051] Each referential expression is parsed manually to form an entity referential equivalence set corresponding to the expression. Further analysis of these entity referential equivalence sets leads to word abbreviation rules and an equivalence substitution dictionary. An entity referential equivalence set refers to the set of words in the context that refer to the same object. For example, in the sentence, "...Apple sued Qualcomm, alleging that it failed to cooperate according to the relevant contract, and Qualcomm has not yet responded...", (Qualcomm, its, Qualcomm) constitutes the equivalence set.

[0052] (1) Equivalent substitution of locative words

[0053] Design replacement words for locative terms based on typical locative terms in the business domain, for example...

[0054] Airport

[0055] Train station – station

[0056] Public transport bus station - bus stop

[0057] Guangdong Province – Yue

[0058] ………

[0059] (2) Rules for abbreviation of locative words

[0060] Administrative division locations: Location name + Province / City name – Abbreviation of province / city (e.g., Guangdong Province is abbreviated as Yue)

[0061] X-Ring Expressway – X-Ring

[0062] ……..

[0063] (3) Rules for abbreviation of titles

[0064] Design abbreviation rules based on typical terms used in the business domain, for example.

[0065] Name + Title – Surname + Title (e.g., Mayor Zhang XX – Mayor Zhang)

[0066] Name + Title – Surname + Title (e.g., Engineer Zhang xx - Engineer Zhang)

[0067] ……..

[0068] (4) Rules for Abbreviation of Names

[0069] Design abbreviation rules for item names based on the typical item names in the business domain, for example.

[0070] License plate number – vehicle number

[0071] Traffic lights – traffic lights

[0072] Public transport vehicles – buses

[0073] Adjective + Vehicle Type Name – Adjective + Vehicle (e.g., Red Dump Truck – Red Vehicle)

[0074] ……..

[0075] (5) Rules for Abbreviating Organization Names

[0076] Design rules for abbreviating organization names based on typical organization names in the business field, for example.

[0077] Administrative region name + People's Government - (Administrative region level + "Government")

[0078] ……..

[0079] (5) Fact description

[0080] Design corresponding pronouns based on the typical factual description types in the business domain, for example.

[0081] Use "the road conditions" in the following text to refer to sentences or phrases describing traffic conditions. Use "the accident" in the following text to refer to sentences or phrases describing traffic accidents. Use "the violation" in the following text to refer to sentences or phrases describing traffic violations.

[0082] …………

[0083] 2. Identify common objects

[0084] (2.1) Segment the dialogue history text of both parties and mark the parts of speech.

[0085] (2.2) Based on the word segmentation results, each word is used as the starting word in turn, and different numbers of subsequent adjacent words are concatenated to form strings of various lengths.

[0086] (2.3) Delete strings that end with adverbs and particles.

[0087] Strings ending with adverbs or particles have incomplete grammatical structures and may have ambiguous semantics, and should be discarded.

[0088] (2.4) Count the frequency of occurrence of semantically similar strings and select the strings whose frequency exceeds the threshold.

[0089] Selecting frequently occurring strings during the conversation and replacing them with shortened versions can help reduce the length of the conversation text.

[0090] (2.5) Use grammatical and semantic methods to identify locative words, names of people, names of organizations and factual descriptions as common knowledge objects. Common knowledge objects are places, titles, names of objects, names of organizations and factual descriptions mentioned by both parties in the dialogue, and they are in the form of words, phrases, sentences or sentence groups.

[0091] The system determines whether a string belongs to a specific category (location, title, object name, organization name, or factual description) based on both grammatical and semantic dimensions, and then performs abbreviation replacement according to the specific category. For example, the string "green dump truck" grammatically conforms to an agglutinative noun phrase and semantically belongs to the category of transportation vehicles; therefore, it is considered a name that can be abbreviated.

[0092] 3. Perform reference and abbreviation replacement.

[0093] Based on word abbreviation rules and equivalent substitution dictionaries, common knowledge objects are expressed using references and abbreviations during robot dialogue, thereby reducing the length of the dialogue text.

[0094] In summary, this invention employs natural language processing technology to discover consensus knowledge by mining historical data of information interaction between robots. This consensus knowledge can be used to reduce the amount of data transmitted in information interaction, thereby reducing the bandwidth requirements of communication transmission lines.

Claims

1. A method for improving the efficiency of information exchange and transmission in robot dialogue, characterized in that, Includes the following steps: Step 1: Construct a dictionary of word abbreviation rules and equivalent substitutions, specifically as follows: (1.1) Automatically discover typical referential expression patterns in massive corpus texts by using language models; (1.2) Each reference expression is parsed through manual analysis to form an entity reference equivalence set corresponding to the reference expression. Then, the various entity reference equivalence sets are analyzed to form word abbreviation rules and equivalent substitution dictionary table. Step 2, identify commonly known objects, specifically as follows: (2.1) Segment the dialogue history text of both parties and mark its parts of speech; (2.2) Based on the word segmentation results, each word is taken as the starting word and concatenated with different numbers of subsequent adjacent words to form strings of various lengths; (2.3) Delete strings that end with adverbs or particles; (2.4) Count the frequency of each string and select the strings whose frequency exceeds the threshold; (2.5) Use grammatical and semantic methods to identify locative words, names of people, names of organizations and factual descriptions as common knowledge objects; common knowledge objects are places, titles, names of objects, names of organizations and factual descriptions mentioned by both parties in the dialogue, and their forms are words, phrases, sentences and sentence groups; Step 3: Based on word abbreviation rules and equivalent substitution dictionary, common knowledge objects are expressed using references and abbreviations during robot dialogue, thereby reducing the length of the dialogue text.

Images