Patent Application
20230245590
The field of the development is that of assisting learning.
More precisely, the development relates to a method for assisting the learning of information by a user of a terminal.
“Information” means in particular, but not exclusively, names of technologies, concepts, places, people, etc., or any other item of information (also called “knowledge” or “skill”) that can be useful to the user.
“Terminal” means in particular, but not exclusively, a personal computer (desktop or laptop), a tablet computer, a personal digital assistant, a smartphone, a workstation, etc., or any other device that a user can use to receive, send or search for content of the text and/or image and/or sound type.
“Content” means in particular, but not exclusively, an email, a message (instant or not), a document, a search (carried out for example with a web browser), a newsfeed of a social network, a content published on a social network, etc.
The development can be applied in numerous fields, for example:
At present, the learning of new knowledge (information, skills) for the employees handling a lot of information at their workstation (for example of the PC type) every day is not carried out in an assisted or proactive manner. When they notice a lack of information or of skills to move forward in an activity, it is up to the employee to do their own research on their workstation or ask for help or a specific training. They can thus lose time in this process of research and learning, or even not find the desired information. Moreover, overwhelmed by the profusion of this new information, they can also forget to search for its meaning.
A close or similar issue exists in numerous fields (field of education, field of personal development, etc.).
The goal of the development, in at least one embodiment, is in particular to overcome these various disadvantages of the prior art.
More precisely, in at least one embodiment of the development, one goal is to provide a technical solution for assisting the learning of information by the user of a terminal.
The goal of at least one embodiment of the development is also to provide such a solution that is simple to implement and easy to use.
Another goal of at least one embodiment of the development is to provide such a solution that allows to limit the computation resources of the computation machine, as well as the network traffic to and/or from the terminal of the user.
A complementary goal of at least one embodiment of the development is to provide such a solution that allows to adapt to both the user and to the information to be learned.
In a specific embodiment of the development, a method is proposed, implemented by a computation machine, for assisting the learning of a plurality of items of information by a user of a terminal. The method comprises, during an encounter of the user with a given item of information, among the plurality of items of information, during a use of the terminal:
Thus, the proposed solution proposes an approach that is indeed novel and inventive, consisting of a method for assisting learning that is implemented in a computation machine. It aims to identify the information that can be not known, or simply forgotten, by the employee (but which can be important, for example in the nature of their profession) by determining an index (or degree) of knowledge for each of these items of information. The proposed solution thus aims to propose to the user elements for understanding as a function of the knowledge index determined. Thus, if the knowledge index is judged to be too low (for example because it is lower than a threshold), elements for understanding are proposed to the user, whereas if the knowledge index is high, no element for understanding is proposed.
Thus, an advantage of the proposed solution is that it allows to limit the computation resources used by the computation machine, as well as the network traffic to and/or from the terminal of the user, since the number of notifications, to propose to the user to access the elements for understanding then to provide these elements for understanding (if the user desires it), is limited (the computation machine automatically selects the information for which it is necessary to propose elements for understanding).
The learning of the user is improved. Moreover, by adapting the elements for understanding as a function of the knowledge index, for example by reducing the size of the chosen element for understanding, this also allows to limit the network traffic. For example, for a value of the index indicating that the given item of information is not known to the user (for example the value 0 or a value close to 0), the element for understanding proposed is a complete initial training, whereas for a value of the index close to a threshold (for example 0.45 if the threshold is equal to 0.5) and indicating that the given item of information is possibly forgotten by the user, the element for understanding proposed is for example a shorter training (or a simple reminder of a definition).
One advantage of the proposed solution is that it is simple to implement since besides the terminal, which the user already has available, a computation machine (optionally the one already present in the terminal) cooperating with a knowledge base (database) suffices.
Another advantage of the proposed solution is that it is easy to use since the user is offered access to elements for understanding, for information that the computation machine itself selected automatically, as a function of the knowledge index determined.
Another advantage of the proposed solution is that it allows to adapt both to the user (since the knowledge index is a function of the content of the knowledge base, which itself depends on the choices of the user to access the proposed elements for understanding or not) and to the item of information to be learned (since the knowledge index is specific to a given item of information; in other words, each of the items of information is associated with its own knowledge index). Thus, the proposed solution allows to take into account the ease or on the contrary the difficulties that each user can encounter in specific areas of expertise (that is to say for specific items of information to be learned).
According to a specific feature, in the case of access, respectively non-access, by the user to said at least one element for understanding, the method comprises a storage, with said entry, of at least one item of context data relative to said access, respectively to said non-access.
Thus, the contents of the knowledge base are richer, which allows to further improve the computation of the knowledge index.
According to a specific feature, the encounter of the user with the given item of information belongs to the group comprising:
Thus, the proposed solution can take into account the large diversity in the encounters that the user can have with a given item of information. It is effective even if the user handles a large quantity of information. The list of types of encounters is not exhaustive.
According to a specific feature, said at least one element for understanding belongs to the group comprising:
In this way, the proposed solution can provide a large diversity in the elements for understanding proposed to the user for their learning of a given item of information. The list of types of elements for understanding is not exhaustive.
According to a specific feature, said at least one item of information data belongs to the group comprising:
The greater the number of items of information data managed (that is to say the more the item of information is specified), the more finely the knowledge index can be computed. “Type of the encounter of the user with the given item of information” means for example a choice between “reading”, “writing” and “search” (non-exhaustive list).
According to a specific feature, the context data belongs to the group comprising:
Just like for the information data, the greater the number of items of context data managed (that is to say the more the item of information is specified), the more finely the knowledge index can be computed.
According to a specific feature, the determination of the knowledge index is a function:
In this way, the knowledge index can be computed easily during a certain number of first iterations of the method for assisting learning and as long as the knowledge base is not sufficiently filled for a computation of index based on a machine learning model (see detail below) to be considered as acceptable.
According to a specific feature, the determination of the knowledge index uses a machine learning model and comprises:
Via the use of a machine learning model (also called “machine learning”), the knowledge index can be computed with good performance and on the basis of several criteria (corresponding to the various items of information and context data).
According to a specific feature, the attributes comprised in the item of entry data belong to the group comprising:
The greater the number of attributes managed in the item of entry data, the more finely the knowledge index can be computed. This list of attributes is not exhaustive.
According to a specific feature, the method comprises a building of the machine learning model, by carrying out a determined number of building iterations, each corresponding to an iteration of the method for assisting learning, and each comprising the following steps:
Thus, the machine learning model can be built via the information and context data stored during certain iterations of the method, while taking into account in particular the access or not by the user to the elements for understanding.
According to a specific feature, the estimation of the knowledge index is equal to:
For example, the first value is “0” and the second is “1”.
According to a specific feature, the building of the machine learning model is carried out again after a predetermined number of iterations of the method for assisting learning and/or at a predetermined frequency.
In this way, the machine learning model can change over time, to improve the learning as the number of iterations of the methods increases, that is to say as the contents of the knowledge base increase.
In another embodiment of the development, a computer program product is proposed comprising program code instructions which, when they are executed by a computation machine, cause the computation machine to carry out the aforementioned method (in any one of its various embodiments).
In another embodiment of the development, a storage medium readable by a computer and non-transient, storing the aforementioned computer program product, is proposed.
In another embodiment of the development, a computation machine configured to carry out the aforementioned method (in any one of its various embodiments) is proposed.
Other features and advantages of the development will appear upon reading the following description, given as an informational and non-limiting example, and the appended drawings, in which:
In all the drawings of the present document, the identical elements and steps are designated by the same numerical reference.
Now, in relation to the flow chart of
The method is implemented by a computation machine (also called “system” in the rest of the description), for which an example of structure is presented below, in relation to
In a step E1, the computation machine examines, via one or more probes, the activity of the user on the terminal, comprising for example the content of the text and/or image and/or sound type that the user has received, sent or searched for. As already mentioned above, “content” means in particular, but not exclusively, an email, a message (instant or not), a document, a search (carried out for example with a web browser), a newsfeed of a social network, a content published on a social network, etc. In this step E1, it is supposed that the rights to access the data of the professional activities of the user locally are authorised in compliance with the General Data Protection Regulation (GDPR).
In a step E2, the computation machine analyses the contents examined and attempts to extract therefrom information that must be the object of learning by the user. The extraction is based for example on referential linguistic expression (named entities) and simple and extended phrases. As already mentioned above, this information (also called “key elements”) is for example names of technologies, concepts, places, people, etc., or any other item of information (also called “knowledge” or “skill”) that can be useful to the user.
In other words, the computation machine detects an encounter of the user with one or more items of information (key elements), during a use of the terminal. “Encounter of the user with a given item of information” means for example the presence of the given item of information in:
In a test step E3, for a given item of information extracted in step E2 (that is to say for an encounter of the user with this given item of information), the computation machine determines whether there is already an entry for the given item of information in a knowledge base 1 (database) aggregating information (key elements) for the user. For this, the computation machine queries the knowledge base 1, as symbolised by the arrow labelled 2.
If no entry for the given item of information exists in the knowledge base (negative response to the test of step E3), the algorithm continues to step E7 in which the computation machine creates in the knowledge base 1 an entry for the given item of information, and stores with this entry at least one item of information data relative to the given item of information and at least one item of context data relative to the encounter.
In step E7, the computation machine stores for example:
Step E7 is followed by a step E8 in which the computation machine proposes to the user to access (at least) one element for understanding the given item of information. In a specific implementation, the element for understanding is chosen according to a value of the knowledge index. “Element for understanding” means for example: a definition of the given item of information, an explanation of the given item of information, a training on the given item of information, help relative to the given item of information, an element for learning (written and/or oral and/or visual) the given item of information, etc.
In a specific implementation, the element for understanding is structured in such a way that it includes a basic item of information that is completed or enriched according to a defined tree structure, the elements of this tree structure being selectable according to the value of the knowledge index.
In another possible implementation, the element for understanding can be structured by different information on size or quantity of data, in a data matrix for example. This matrix is built while taking into account for example characteristics of duration of reading of this element for understanding or of complexity.
Thus, according to the value of the knowledge index, an element for understanding having a longer or shorter duration of reading or a greater or lesser complexity can be selected in this matrix.
Thus the lower the knowledge index, the more the element for understanding must be complete, and thus have a greater reading time and/or a larger size and vice versa when the knowledge index is high.
If an entry for the given item of information already exists in the knowledge base (positive response to the test of step E3), the algorithm continues to step E4 in which the computation machine stores in the knowledge base, with the existing entry, at least one (other) item of context data relative to the encounter.
Then, in a step E5, the computation machine computes an index of knowledge (by the user) of the given item of information, according to the contents of the knowledge base.
For the computation of the knowledge index, one and/or the other of two methods is for example used, according to the number of iterations of the method already carried out before the current iteration. For example, for the N first iterations (in a specific implementation, N=1000), a first method based on a forgetting curve is used, and for the following iterations, a second method based on a machine learning model is used.
The first method comprises for example a computation of the knowledge index according:
The second method comprises for example the following steps to compute the knowledge index:
In a specific implementation, the item of entry data comprises the following attributes:
Step E5 is followed by a test step E6, in which the computation machine compares the knowledge index to a predetermined threshold. If the knowledge index is greater than or equal to the threshold, the algorithm goes back to step E1, for a new iteration of the method. If the knowledge index is lower than the threshold, the algorithm continues to step E8 already explained above (proposition of access to at least one element for understanding).
Step E8 is followed by a test step E9, in which the computation machine determines whether the user accessed the element for understanding.
In the case of access, the algorithm continues to step E10 in which the computation machine stores in the knowledge base (as symbolised by the arrow labelled 3), with the existing entry, at least one other item of context data, relative to the access. After step E10, the algorithm goes back to step E1, for a new iteration of the method.
In the case of non-access, the algorithm continues to step E11 in which the computation machine stores in the knowledge base (as symbolised by the arrow labelled 4), with the existing entry, at least one other item of context data, relative to the non-access. After step E11, the algorithm goes back to step E1, for a new iteration of the method.
The building of the model comprises a determined number M of building iterations, each corresponding to one of the iterations of the method for assisting learning of
In a specific implementation, the M building iterations correspond to the N first iterations of the method of
In a step 31, the computation machine generates an item of entry data as defined above (see step E5 of
In a step 32, the computation machine determines an estimation of the knowledge index, according to the access or the non-access by the user to the element for understanding (see test step E9 of
In a step 33, the computation machine provides to the machine learning model the item of entry data, accompanied by the known result (defined as the estimation of the knowledge index, computed in step 32).
In a test step 34, the computation machine determines whether the number M of building iterations has been carried out. If no, the algorithm goes back to step 31, for a new building iteration. If yes, the algorithm continues to the end step 35.
The building method of
This structure comprises a random-access memory 42 (for example a RAM memory), a read-only memory 43 (for example a ROM memory or a hard disk) and a processing unit 41 (equipped for example with a processor, and controlled by a computer program 430 stored in the read-only memory 43). Upon initialisation, the code instructions of the computer program 430 are for example loaded into the random-access memory 42 before being executed by the processor of the processing unit 41.
This
In the case of an implementation in the form of a reprogrammable computation machine, the corresponding program (that is to say the sequence of instructions) can be stored in a storage medium removable (for example such as a diskette, a CD-ROM or a DVD-ROM) or not, this storage medium being partly or totally readable by a computer or a processor.
The present application example attempts to resolve the following issue: how to identify/detect an item of information not known to or poorly known (capable of being forgotten) by the employee with regard to several criteria (for example, the period of time between a determined number of appearances (encounters) of this same item of information in their activities, the nature or the field associated with said item of information . . . ).
For example, an employee of a business can receive an electronic message (email) likely to contain: “For this, you must make an RCE”. If the acronym RCE (meaning “Request for Computer Equipment”) is not an item of information known to the employee, they are very likely to not be capable of carrying out the desired action. The method for assisting learning (see
The method aggregates in a database information forming the object of the learning (key elements) such as: names of technologies, concepts, places, people, or any other item of information that the employee encounters in their daily activities (emails, conversations on a messaging service (instant or not), documents . . . written or read) during a period more or less long and configurable according to their preferences or their profession.
The dates of encounter of the item of information by the employee are also listed in this knowledge base. And the nature of the encounter context in particular: is it in incoming emails? Is it a manual search for the item of information by the employee? Has the definition (forming an element for understanding) of this item of information proposed by the system (also called “computation machine” above) been read?
This context data is used to compute the index of knowledge of this item of information by the employee.
By analysing the entries of the employee as they arise such as the elements (contents) that they receive: emails, conversations, new read documents, the system detects the information contained in these entries and verifies the presence of this information in the knowledge base of the employee. If an item of information detected in one of the contents is not present in the knowledge base: the system considers that this item of information is potentially new, and thus not known to the employee.
If the item of information is already present in the knowledge base, the system computes a knowledge index. This index is based on several criteria, for example: the period of time between two appearances of the item of information (which can suggest forgetting if this period of time is long), the field associated with this item of information (network, project management, new technologies, etc.), the nature of the item of information (tools of the IS, contact, organisation, etc.), the context of appearance of the item of information, etc. The index is computed on the observation of the knowledge base of the employee and, in a specific embodiment, on the observation by the system of requests by the employee for additional information (or on the contrary the absence of requests for additional information) for information of a similar field or nature. The knowledge index is thus specific to the item of information and specific to the user.
When the method is started, the knowledge base can be initialised by information to which the employee has given access over a given period. Then the index will be based on average memorisation hypotheses, such as those that are visible on the Ebbinghaus Curve.
A) Let us consider the item of information (acronym) “APN” that the employee reads in an email (step E1). It is not present (step E3) in the knowledge base that contains information corresponding for example to a month (initialisation on the basis of information coming from the history of the emails, instant or other messages).
The system thus stores (step E7) the following data (non-exhaustive list) in the knowledge base:
The system displays (E8) a proposition of access to the definition of this acronym (item of information) to the user.
If the user clicks on the definition (that is to say accesses the element for understanding), the system considers that the definition was useful to them and adds (E10) into the knowledge base the following new context data (in particular an item of context data indicating “Item of information not known”), which will allow to produce a specific learning model associated with the employee:
If the user does not click on the definition (that is to say does not access the element for understanding), the system extrapolates and considers that the user knows this item of information. It adds (E11) into the knowledge base the following new context data (in particular an item of context data indicating “Information known”), which will allow to produce the specific learning model associated with the employee:
Wording: APN
B) It is supposed that one month later, the user reads the acronym (item of information) “APN” in an instant message (IM). The system detects (E3) that the item of information exists in the knowledge base and adds (E4) into the knowledge base the following new context data:
At this stage, the knowledge base contains too little information and data to extrapolate and compute the index on the knowledge that the system has of the employee. It thus computes (E5) the knowledge index on the basis of the Ebbinghaus curve.
The item of information (acronym “APN”) has not been seen for a month: during the display of the definition of the item of information. The system considers that, in these conditions (period of time of one month), 20% of the item of information has been retained: the index is thus assigned at 0.2. The system once again displays the definition (E8).
If the user clicks on the definition (that is to say accesses the element for understanding), the system considers that the definition was useful to them and adds (E10) into the knowledge base the following new context data (in particular an item of context data indicating “Item of information not known”):
If the user does not click on the definition (that is to say does not access the element for understanding), the system extrapolates and considers that the user knows this item of information. It adds (E11) into the knowledge base the following new context data (in particular an item of context data indicating “Item of information known”):
Wording: APN
C) If the user themself searches for the definition of the item of information (acronym “APN”) (E1), the system detects it (E2). If the item of information is not present (step E3) in the knowledge base, the system thus stores (E7) the following data in the knowledge base:
Wording: APN
The system displays (E8) a proposition to access the definition of this acronym (item of information) to the user. See paragraph A above for the rest (according to whether the user clicks on the definition or not, passage to step EH) or E11).
The knowledge base will thus be filled with information and data specific to the user and allow the production of a model capable of assigning a knowledge index to each of the items of information.
The data of the knowledge base is analysed and allows the production of entry data, for a conventional process of building (formation) of the machine learning model.
Each item of entry data is accompanied by the associated known result (knowledge index is equal to “1” if the item of information is known or to “0” if the item of information is not known).
The building algorithm then identifies trends in these entry data and results, which will match the attributes of the entry data with the result (target, that is to say the value of the knowledge index to be predicted). And it provides at the output a machine learning model that captures these trends.
The attributes (characteristics) to take into account to produce an item of entry data of the machine learning model are for example (non-exhaustive list):
Wording
The associated result (response) is: knowledge index=1 (item of information known) or 0 (item of information not known).
The associated result (response) is: knowledge index=1 (item of information known).
The knowledge index is in the method the result, that is to say the qualitative variable to be predicted by learning.
Each item of entry data produced upon analysis of the knowledge base is for example vectorised. Each value of one of the attributes (characteristics) of an item of entry data is turned into an item of digital data to produce a vectorised item of entry data. Then via artificial intelligence (AI) algorithms of the classification type, the model will allow to obtain the knowledge index, a real value varying from 0 (which means that the item of information is not known) to 1 (which means that the item of information is known) for a new vectorised item of entry data allowing to answer the question (with values inputted for the indications A to L): “What is the knowledge index of an item of information with the wording A, from the field B, of nature C, of type D, for which the last appearance (encounter) took place between E and F months, for which the number of appearances in reading took place G times, in writing H times, in a context of appearance of more than I sentences in reading and J sentences in writing, for which the definitions were read K times the month preceding the last appearance . . . ?”. The building algorithm selected depends for example (“deep learning”, “LSTM neural network” . . . ) on results of tests on the entry data and can change over time.
| Number | Date | Country | Kind |
|---|---|---|---|
| FR2006657 | Jun 2020 | FR | national |
This application is filed under 35 U.S.C. § 371 as the U.S. National Phase of Application No. PCT/FR2021/051095 entitled “METHOD FOR ASSISTING A USER OF A TERMINAL TO LEARN A PLURALITY OF ITEMS OF INFORMATION” and filed Jun. 17, 2021, and which claims priority to FR 2006657 filed Jun. 25, 2020, each of which is incorporated by reference in its entirety.
| Filing Document | Filing Date | Country | Kind |
|---|---|---|---|
| PCT/FR2021/051095 | 6/17/2021 | WO |
