SYSTEM AND METHOD FOR EXCHANGE OF SEMANTIC DATA

Abstract

The invention relates to a semantic communication system comprising at least one transmitter (11) of a message M, a receiver (13) of said message M after transmission via a communication channel (12), characterized in that it comprises at least one message content interpretation module (14) and an explanation module (15) of the parameters used by the message interpretation module. Figure for the abstract: Fig. 2

Description

Title of the invention: SYSTEM AND METHOD FOR EXCHANGE OF SEMANTIC DATA

[0001] The invention relates to a system and a method used for exchanging semantic data between a source and a destination, configured to provide information on the semantics chosen during the transmission and / or reception of the semantic data (an explanation of the semantics used by the sender and / or the receiver). This information on the semantics can, in particular, be used to optimize the final use of the semantic data.

[0002] The invention is notably used in semantic communication systems for the transmission of semantic messages and in semantic communication systems in which the purpose of the communication is known or estimated at the level of a receiver.

[0003] In 1948, Shannon and Weaver classified communication into three levels: (i) the transmission of symbols, (ii) the semantic exchange of transmitted symbols and (iii) the effect of the exchange of semantic information.

[0004] Over the past seventy years, based on the Shannon and Weaver theory, communications have focused primarily on how to accurately and efficiently transmit data symbols between a transmitter and a receiver, freeing themselves from possible disturbances introduced by the propagation channel.

[0005] The development of next-generation communication networks, for example 5G networks, and beyond, leads to imagining solutions that allow the selection of semantic data useful for communication and means of processing this data, so that it is understandable by the recipient, regardless of the final processing.

[0006] Semantic inadequacy can be caused by contradictions at the knowledge base level, semantic infidelity, approximation in semantic communication or exchange of semantic data, computation and / or caching, cultural differences between source and destination, inadequate interpretation of the purpose of communication between source and destination, or other factors.

[0007] In the following description, the term "semantic interpretation module" or "semantic explanation module" refers to a module whose function is, in particular, to provide an explanation of the choice made by the message interpretation module to "decode" (in the semantic sense) the message. semantics received at the level of a receiver. For example, the semantic explanation module will provide the parameters used by the interpretation module to interpret the semantic message.

[0008] The idea of ​​the present invention is to propose a new semantic data exchange system and a new method that will provide explanatory information associated with a decision made by a message interpretation module to "decode" the message semantically. In particular, the invention makes it possible to explain the decision / choice made by the message interpretation module to interpret the message so that the semantic data is understandable to the recipient and / or to a message generation module at the sender's end. This explanation can then be used to reduce the effect of semantic inadequacy on the effectiveness of the tasks intended by the recipient or the sender.

[0009] The invention relates to a semantic data exchange system comprising at least one semantic data transmitter M including a semantic data generator, a receiver of said semantic data M after transmission via a transmission channel characterized in that it includes at least one semantic data content interpretation module and one semantic parameter explanation module used by the message interpretation module.

[0010] The semantic data content explanation module is located at the receiver level; said semantic data content explanation module receives the message interpreted by the semantic data interpretation module and generates information.

[0011] The module for explaining the content of semantic data can also be placed at the level of the sender and generate a set of parameters specific to the choice of the construction of semantic data.

[0012] According to another embodiment, the module for explaining the content of the semantic data is centralized outside the communication system.

[0013] The system may further include a semantic data interpretation correction module and / or a module configured to control the "quality" of the semantic data interpretation result.

[0014] The data exchange system is a communication system in which semantic data are semantic messages.

[0015] The invention also relates to a method for exchanging semantic data within a semantic data exchange system according to the invention, characterized in that it comprises at least the following steps:

[0016] - Interpret the content of semantic data in order to extract information explanatory semantics regarding the semantics used for said semantic data,

[0017] - Use said extracted semantic information.

[0018] The process will interpret the semantic data received at the receiver level to improve semantic understanding before use of the semantic data by a recipient.

[0019] Semantic data can also be interpreted at the sender's level in order to improve the content of the semantic data to be transmitted according to the receiver's expectations.

[0020] Semantic data is presented, for example, in the form of a semantic message.

[0021] Other features, details and advantages of the invention will become apparent from the description made with reference to the accompanying drawings given by way of illustrative and non-limiting example and which represent, respectively:

[0022] [Fig. 1], a general diagram of a semantic communication system according to the prior art,

[0023] [Fig.2], an illustration of the system architecture according to the invention comprising a semantics extraction module and a module for selecting the semantics retained during message transmission,

[0024] [Fig.3], an example of a communication system comprising a transmission channel to the transmitter.

[0025] Figure 1 illustrates an example of a prior art system comprising a message source 9 connected to a transmitter 10 comprising a message generator 11 that provides a message x = g(w), where g is a message generation function that takes as input the message w generated by the message source 9 and information from the BGKS knowledge base. The message x is transmitted via a wireless communication channel 12 to a message interpretation module 14 included in a receiver 13. The message interpretation module 14 is configured, in particular, to interpret the content of the message y transmitted by the communication channel in order to make it readable by the recipient. The interpreted message w' = g(y) is then transmitted to a destination 15, g being an interpretation function. The system also includes a BGKD knowledge base at the recipient or receiver.These knowledge bases are more commonly known by the English term "Background". They can contain contextual information, elements known a priori. These elements will be used in particular to correctly interpret semantic data so that it is understandable by the sender and / or the recipient in order to be used.

[0026] Figure 2 illustrates an example of an embodiment of the invention comprising a module 20. In this example, the semantic explanation module is located at the receiver of the system containing the interpretation module. 14 of the semantic message M. The elements common to the system described in [Fig.1] bear the same references.

[0027] Without departing from the scope of the invention, the semantic explanation module 20 can also be located at different points in the communication system, at the transmitter or in a centralized location such as a "cloud". In the latter case, the semantic explanation module will receive the message from the source or the message from the interpretation module or both.

[0028] The semantic explanation module 20 is configured to provide additional information on the content of the interpreted message, i.e., how the message is generated, how the message was interpreted. For example, the Bayesian inference method may be used.

[0029] One of the objectives of the system according to the invention is not to reproduce exactly or approximately the transmitted message, but rather to obtain information on how the message interpretation module interpreted the message in order to make it readable and understandable.

[0030] Whereas in a classical communication system, a message source consists mainly of a sequence of symbols, in the case of semantic communications, a message will be syntactically considered as a sequence of symbols - an expression composed using symbols in the source language (modulation constellation, specific coding, sequences, etc.) - the meaning of this message could be misinterpreted at the receiver level due to semantic noise.

[0031] An example of a conversation between three people is given to better illustrate the object of the invention. Considering the conversation between three people:

[0032] Giulio: "Are you home this weekend?"

[0033] Livio: "Emilio, Giulio is asking if you're free to play with us this weekend?"

[0034] Emilio: "No, I'm not available on Saturdays and Sundays."

[0035] Livio serves as a semantic channel between Giulio and Emilio. Livio does not transmit precisely the original message from Giulio to Emilio. However, Emilio is still able to understand the original meaning of Giulio's message.

[0036] In the case of semantic transmission, Emilio understands the meaning of Giulio's message, therefore there is no failure in the transmission of the message.

[0037] The semantic explanation module according to the invention has, in particular, the function of extracting the semantics on the interpretation of the source message and / or the received message, the parameters which were taken into account to decode the message in order to make it readable by the recipient, for example.

[0038] Information on the choice of semantics used for reading a message can be exploited when the interpretation criterion evolves according to environmental conditions.

[0039] For example, an object in an image can be interpreted by taking into account the colors contained in the image and / or by taking into account the shapes. The semantic interpretation module can use the semantic explanation information to adjust its choice of interpretation of the message content (the chosen interpretation function, the priority, the acceptable errors, etc.). The same approach can be used by the semantic message generator located at the sender.

[0040] The information can also be used to implement mechanisms that adapt the encoding / decoding parameters of the semantic message content at the source or receiver level. After a misinterpretation of a semantic message is detected at the receiver level, for example, the receiver can decide to adapt the semantics used at the sender and / or receiver level to achieve a correct interpretation of the semantic message by the recipient. This explanatory information in the choice of interpretation can be used to improve the recipient's understanding of the message.

[0041] The information generated by the message interpretation explanation module can also be used to enrich the knowledge bases at the sender level and / or at the receiver level.

[0042] The system may also include an information aggregation module 30 ([Fig.3]) which will allow the collection of interpreted semantic messages before deciding whether the received semantic messages are capable of fulfilling their function, whether they are capable of allowing the recipient to carry out the actions planned upstream of the exchange system.

[0043] This aggregation module 30 also makes it possible to associate a confidence value or interpretation score with the interpreted semantic messages, i.e., a probability that the interpreted semantic message is similar or equivalent to the emitted semantic message. This aggregation module is, for example, placed between the semantic message interpreter module and a message interpretation correction module 31. The latter is configured to correct any erroneous interpretation of the message.

[0044] This aggregation module aggregates the differences between all models of a message and those of the "true" state. After each received message, the receiver accumulates the interpreted semantic messages and performs an interpretation check on the message sequence. An interpretation score is associated with the aggregated interpretation. This makes it possible, in particular, to associate a level of confidence in the mutual understanding of the semantic message between the source and the destination.

[0045] According to one embodiment, the semantic communication system may also include a module 32 configured to monitor the quality of the interpretation of the semantic message, i.e., whether the message was interpreted correctly or insufficiently to be understood. This module's function is, in particular, to draw a conclusion about the understanding of the received message based on the interpreted message received and the knowledge base of the receiver and / or the sender when such information is available to the latter. This will allow for the estimation of a confidence level regarding the mutual understanding between the source and the destination, and the communication objective.

[0046] If the interpretation is correct, then the system can continue to receive the next semantic message. In case of incorrect interpretation, a semantic transmission mechanism can be triggered.

[0047] The following example of implementation of the invention is given by way of illustration for a system which includes a semantic message interpretation module at the sender level and a semantic message explanation module 20 at the receiver level 13 ([Fig.2]).

[0048] At the transmitter 10, the generation module 11 generates a semantic vector describing the parameters or semantics it used to generate the semantic message.

[0049] For example, the semantic vector provides the explanation based on the BGKS knowledge base at the source (absent, partial, uncertain, fully available). This may be a "risky message generation", unconventional for the coder, based on partial information, in the presence of dynamic parameters in the context of communication.

[0050] The semantic vector may also include an explanation indicating "how the elements of the recipient's or receiver's knowledge base were taken into account" to generate the semantic message so that it is understandable by the recipient and / or the sender.

[0051] The semantic interpretation or explanation module can also generate explanatory information explaining the influence of data source filtering (generating a semantics that encompasses different data from a source, different data from different sources). For example, the explanation could be how the data is merged at the source and the compression parameter at the message generation level.

[0052] In the field of semantic communication, the message content must be interpreted in a comprehensible manner. The semantic interpretation module at the receiver level can use knowledge graphs for the detection of semantic similarity or measured equivalences between the message content The interpreted message and the gap between the messages that are expected to be received are taken into account, given the BGKD knowledge base at the receiver level and potentially a complete or partial knowledge of the BGKD knowledge base at the sender (message generator). This module is supplemented by information generated by the semantic explanation module, which is configured to generate information about the parameters used to semantically decode the message.

[0053] For example, this information can explain:

[0054] - How the elements of the knowledge bases at the sending and receiving stages have influenced the interpretation of the message,

[0055] - How the noise present in the transmission chain influenced the interpretation of the message. It will be possible to distinguish the origin of the noise: noise from the transmission channel, semantic noise, noise from the knowledge base,

[0056] - If the semantic equivalences and the validity of the probabilities, scores of Trust is possible.

[0057] - The detection of "inconsistency" between the target purpose of the communication between the source and destination, or only an ambiguity noted in the understanding of the purpose of the communication upon reception.

[0058] The semantic message interpretation module at the receiver level that receives the message w' can suggest an interpretation vector and add to each interpretation vector a confidence value, for example, a probability value that the semantics of the received message is identical to the semantics of the transmitted message.

[0059] The semantic message interpretation module assigns probability values ​​to the interpreted semantic messages to determine the likelihood that the received message is true according to a knowledge base at the receiver level. If this information is available at the sender level, it is also possible to take into account the sender's knowledge base in whole or at least partially.

[0060] The description just given applies to any system comprising the presence of one or more senders (unit that can transmit a message to an identifiable or unidentifiable destination) and one or more destinations (mobile terminals, smart mobile phones, mobile Edge hosts, etc.).

[0061] The invention has in particular the advantage of offering additional explanatory information on the parameters, on the semantics, used for the development or decoding of a semantic message in a semantic communication system or more generally on the exchange of semantic data.

[0062] The present invention enables a new type of semantic adaptive mechanism that can be fed by an explanation of the choices and decisions made by the message interpretation module, in order to improve these choices and the decisions to be made to optimize the "reading" of the semantic message.

Claims

Demands

1. A communication system comprising at least one transmitter (10) and one receiver, the system being configured to transmit semantic data forming a message to said receiver via a wireless communication channel (12), the transmitter comprising a generator (11) configured to generate said message from a source connected to the transmitter, the transmitter and the receiver each comprising a knowledge base, characterized in that it comprises at least one interpretation module (14) configured to perform semantic decoding of the message received by the receiver taking into account at least the knowledge base of the receiver, and in that the system comprises at least one explanation module (20) configured to determine semantic parameters used by the message interpretation module to decode the message.

2. System according to claim 1, characterized in that the semantic data content explanation module (20) is disposed at the level of the receiver (13).

3. System according to claim 1, characterized in that the semantic data content explanation module (20) is arranged at the level of the emitter (10).

4. System according to claim 1, characterized in that the module for explaining the content of semantic data is centralized outside the communication system.

5. System according to any one of claims 1 to 4, characterized in that it further comprises a semantic data interpretation correction module (31) configured to correct an erroneous interpretation of the message by the interpretation module (14).

6. System according to any one of claims 1 to 4, characterized in that it further comprises a module (32) configured to control the "quality" of the result of interpreting semantic data from the interpreted message and the knowledge base of the receiver and / or the sender.

7. A system according to any one of the preceding claims, characterized in that it comprises an information aggregation module (30) for associating interpreted messages with an interpretation score corresponding to the probability that the interpreted message is similar or equivalent to the message sent.

8. System according to any one of the preceding claims, characterized in that the parameters provided by the explanation module (20) include the elements of the sender's and receiver's knowledge bases that influenced the interpretation of the message.

9. System according to any one of the preceding claims, characterized in that the parameters provided by the explanation module (20) include information on the influence of noise present in the transmission chain on the interpretation of the message.

10. A system according to any one of the preceding claims, characterized in that the explanation module (20) further functions to extract the semantics on the interpretation of the message emitted by the source, and to use it to implement mechanisms that adapt the encoding parameters of the content of the semantic message at the source level.

11. A data exchange method implemented in a receiver within a communication system according to any one of claims 1 to 10, characterized in that it comprises at least the following steps: - extracting semantic information on the interpretation of said message, - using said extracted semantic information.

Images