College office administrative document automatic management method and system

Abstract

The invention discloses an automatic management method and system for university office administrative documents, and the method comprises the steps: constructing a three-dimensional tensor containing a document type, a process state and time sensitivity, generating an initial state tensor, calculating a circulation probability through a random process, and supporting conventional circulation and emergency jump; obtaining a to-be-processed document, extracting a feature vector, and mapping the feature vector to a state space; calculating a state transition probability based on a circulation probability function, and triggering a jump mechanism to adjust a path in combination with time sensitivity; and determining the next state and calculating the user permission in real time. Compared with a traditional method, the method has the advantages that document processing time can be saved through dynamic path optimization, permission conflicts are avoided, auditing efficiency is improved, cross-department cooperation efficiency is improved, management efficiency and safety are remarkably improved, and the method adapts to complex requirements of university administrations.

Description

Technical Field

[0001] The present invention relates to the technical field of document management, and in particular to an automated management method and system for administrative documents in university offices. Background Art

[0002] With the continuous improvement of the level of informationization of college administration, office document processing has gradually shifted from traditional manual operations to automated systems to improve efficiency and standardization. The existing document management system has realized the digital storage and process management of documents to a certain extent, such as recording document status through electronic forms, setting fixed approval paths, etc. Patent CN102385725A discloses a document management system and method based on workflow drive, which subdivides documents into work units, establishes a mapping relationship between work units and process activities, and realizes the push and monitoring of documents based on dependency and constraint analysis (such as uniqueness and integrity constraints), supporting the fine-grained management of documents in enterprises. However, there are significant deficiencies in this patent scheme. Its circulation path depends on the predefined activity sequence and static mapping, and it is difficult to flexibly respond to sudden demands in college administration. For example, when the approval of scientific research contracts needs to skip links due to time constraints, the system cannot be adjusted dynamically, resulting in processing delays. In addition, the scheme does not fully optimize the efficiency of cross-departmental collaboration, and it is easy to reduce the overall efficiency due to collaboration bottlenecks when multiple departments participate. Although its monitoring function supports status tracking, the authority control is still statically configured, lacking the ability to dynamically adjust with changes in process and time, which increases authority conflicts and security risks. Summary of the invention

[0003] Purpose of the invention: In order to overcome the deficiencies in the prior art, the present invention provides an automated management method and system for administrative documents in university offices that can dynamically optimize circulation paths, improve collaboration efficiency, and realize real-time authority control.

[0004] Technical solution: To achieve the above purpose, the automated management method of administrative documents in university offices of the present invention comprises:

[0005] Constructing a 3D tensor As the state space of document management, dimension N is the document type space, dimension M is the process state space, and dimension T is the time sensitivity parameter space; the initial state tensor is generated by inputting document template, process specification and time constraint table where α ijk is the document flow probability, β ijk is the department synergy coefficient, γ ijk is the dynamic weight of the permission; the subscripts i, j, and k correspond to the specific indexes of the three dimensions in the state space respectively;

[0006] Get the document to be processed and extract the feature vector Map the document to the corresponding coordinates (i, j, k) in the state space; where s represents the current system state, Φ(s) represents the feature mapping of s in the state space, s′ represents the target state, and Ψ(s′) represents the feature prediction of s′ in the state space;

[0007] The flow probability of the document from the current state s to the target state s′ is calculated based on the flow probability function, and when the time sensitivity parameter T exceeds the threshold, the Lévy jump mechanism is triggered to adjust the flow path of the document accordingly; wherein the flow probability function is:

[0008]

[0009] Where: P(s→s′) is the probability of going from the current system state to the target state s′; Z is the normalization factor, and W is the weight matrix generated by adversarial training;

[0010] Based on the flow path, the next state s1 is determined, dynamic permission control is performed, and the access rights of each user to state s1 are calculated in real time:

[0011]

[0012] Among them, Access(s1,u) represents the quantitative value of whether user u has access rights in document state s1; u represents the vector or identifier of user attributes; K is the total number of permission features; π k (s1) is the k-th level authority characteristic function, which is related to γ ijk Related; k is the original value of user u on the kth feature, will u k Mapped to a probability form; θ is the time decay factor.

[0013] Furthermore, when generating the initial state tensor, the document flow probability α ijk Calculated by Lévy flight random process, the specific formula is:

[0014]

[0015] Where: μ∈(1,2] is the stability index; r is the number of process steps; Γ(1+μ) is the value of the gamma function at (1+μ), which is used to standardize the Lévy distribution;

[0016] When the value of the time sensitivity parameter T exceeds the preset threshold, the value of r is increased to recalculate the corresponding α ijk , and according to α ijk The non-zero characteristic of triggers long-distance state jumps, reducing intermediate links.

[0017] Furthermore, the permission dynamic weight γ ijk Updated by the hypergraph neural network, the update formula is:

[0018] γ t+1 =GAT(γ t ,A);

[0019] Where: γ t is the authority weight at the current time t; γ t+1 is the updated authority weight at the next time (t+1); GAT is the graph attention layer; A is the hypergraph adjacency matrix, which represents the multidimensional relationship between departments, users, and documents.

[0020] Furthermore, the departmental synergy coefficient β in the three-dimensional tensor ijk When it is lower than the preset threshold, the quantum annealing machine is called to solve the Hamiltonian H to generate the optimal department combination; wherein the Hamiltonian H is solved based on the following formula:

[0021]

[0022] in, represents the contribution of the ath department in the collaborative combination v, v is the utility function of the department set, and measures the collaborative effect; A is the set of all departments; Q is the department subset, and |Q| is the number of departments in the subset; is the Paula-Z operator of the ith department; represents the Paula-X operator between the a-th department and the b-th department; J is the collaboration intensity.

[0023] An automated management system for administrative documents in university offices, the system comprising:

[0024] Building blocks for constructing three-dimensional tensors As the state space of document management, dimension N is the document type space, dimension M is the process state space, and dimension T is the time sensitivity parameter space; the initial state tensor is generated by inputting document template, process specification and time constraint table where α ijk is the document flow probability, β ijk is the department synergy coefficient, γ ijk is the dynamic weight of the permission; the subscripts i, j, and k correspond to the specific indexes of the three dimensions in the state space respectively;

[0025] Extraction module, which is used to obtain the document to be processed and extract the feature vector Map the document to the corresponding coordinates (i, j, k) in the state space; where s represents the current system state, Φ(s) represents the feature mapping of s in the state space, s′ represents the target state, and Ψ(s′) represents the feature prediction of s′ in the state space;

[0026] The calculation module calculates the flow probability of the document from the current state s to the target state s′ based on the flow probability function, and when the time sensitivity parameter T exceeds the threshold, triggers the Lévy jump mechanism to adjust the flow path of the document accordingly; wherein the flow probability function is:

[0027]

[0028] Where: P(s→s′) is the probability of going from the current system state to the target state s′; Z is the normalization factor, and W is the weight matrix generated by adversarial training;

[0029] The permission control module is used to determine the next state s1 based on the flow path, perform dynamic permission control, and calculate the access rights of each user to the state s1 in real time:

[0030]

[0031] Among them, Access(s1,u) represents the quantitative value of whether user u has access rights in document state s1; u represents the vector or identifier of user attributes; K is the total number of permission features; π k (s1) is the k-th level authority characteristic function, which is related to γ ijk Related; k is the original value of user u on the kth feature, will u k Mapped to a probability form; θ is the time decay factor.

[0032] Beneficial effects: The method and system for automated management of administrative documents in university offices of the present invention have the following beneficial effects:

[0033] (1) The constructed three-dimensional tensor plays the role of a core data processing and coordination hub in the university document management system. By constructing a three-dimensional tensor state space and combining the flow probability function with dynamic permission control, the efficiency and security of document management are significantly improved. Compared with the traditional method, the present invention optimizes the flow path through feature vector mapping and probability calculation, and performs permission control based on the flow path. The permission control adopts multi-dimensional feature matching and time decay mechanism to reduce permission conflict events, improve processing efficiency, and ensure that access rights are highly matched with status requirements.

[0034] (2) Through the random process calculation formula, when r is small, the α value is high, allowing documents to flow along the standard path. When the time sensitivity T exceeds the preset threshold, the Lévy jump mechanism is triggered to directly adjust the document status, which can regulate the burstiness of document flow and adapt to the non-uniformity of tasks in university administration.

[0035] (3) Real-time update of γ through the hypergraph neural network ijk , so that permission control adapts to process status and time changes, which is better than the static permission model; time decay and dependency capture mechanisms ensure that permissions are only valid when necessary, reduce permission conflicts and abuse risks, automatically associate related permissions, reduce manual configuration, and improve administrative efficiency.

[0036] (4) It can dynamically adapt to changes in administrative processes, demonstrate efficiency and flexibility, and provide an efficient optimization solution for university office document management. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 This is a flowchart of the automated management method for administrative documents in university offices;

[0038] Figure 2 This is a schematic diagram of the composition of the automated management system for administrative documents in university offices. DETAILED DESCRIPTION

[0039] The present invention will be further described below in conjunction with the accompanying drawings.

[0040] like Figure 1 The automated management method for administrative documents in university offices shown in the figure comprises the following steps S101-S104:

[0041] Step S101, construct a three-dimensional tensor As the state space for document management, dimension N is the document type space, dimension N is the process state space, and dimension T is the time sensitivity parameter space; dimension N, also known as the document type space, manages different document types, such as teaching archives, scientific research contracts, personnel records, etc., and specifically distinguishes features through vector encoding; dimension M and the process state space track the status of documents in the life cycle, such as drafting, approval, archiving, and abolition, and each state is associated with operation permissions and flow rules; dimension T, also known as the time sensitivity parameter space, dynamically records time constraints, that is, deadlines, processing time, etc., triggering time-driven automated operations, such as overdue reminders, process acceleration, etc.; the initial state tensor is generated by inputting document templates, process specifications, and time constraint tables where α ijk is the document flow probability, β ijk is the department synergy coefficient, γ ijkis the dynamic weight of authority; the subscripts i, j, and k correspond to the specific indexes of the three dimensions in the state space respectively; for example, i=1, 2, and 3 can be set to represent the document types as teaching archives, scientific research contracts, and personnel records respectively; j=1, 2, 3, and 4 can be set to represent the document status as drafting, approval, archiving, and abolition respectively; k=1, 2, and 3 can be set to represent the processing time as regular, urgent, and overdue respectively.

[0042] Step S102: Obtain the document to be processed and extract the feature vector Map the document to the corresponding coordinates (i, j, k) in the state space; where s represents the current system state, Φ(s) represents the feature mapping of s in the state space, s′ represents the target state, and Ψ(s′) represents the feature prediction of s′ in the state space; the document to be processed may be in the form of a document scan, a spreadsheet, a meeting recording, etc.;

[0043] Step S103, based on the flow probability function, the flow probability of the document from the current state s to the target state s′ is calculated, and when the time sensitivity parameter T exceeds the threshold, the Lévy jump mechanism is triggered to adjust the flow path of the document accordingly; wherein the flow probability function is:

[0044]

[0045] Where: P(s→s′) is the probability of going from the current system state to the target state s′; Z is the normalization factor, W is the weight matrix generated by adversarial training; Z = ∑ s′∈S exp(-‖Φ(s)-Ψ(s′)‖ 2 W), is the exponential sum of all possible target states, which is used to ensure that ∑ s′∈S P(s→s′)=1.

[0046] Step S104, determining the next state s1 based on the flow path, performing dynamic permission control, and calculating the access rights of each user to the state s1 in real time:

[0047]

[0048] Among them, Access(s1,u) represents the quantitative value of whether user u has access rights in document state s1. The larger the value, the higher the authority. It may be possible to determine whether authorization is granted through a threshold (such as Access(s1,u)>τ, τ is the threshold); u represents the vector or identifier of user attributes, for example, u=(department, role, authority level), u provides the user's identity information, which is used to match the authority requirements of state s; K is the total number of authority features, indicating the authority dimension or level that needs to be evaluated; π k(s1) is the k-th level permission feature function, which returns the score of the k-th permission feature according to the state s1. Its value is in the range of [0,1] and is related to γ ijk Related; k is the original value of user u on the kth feature (such as department matching degree), will u k Mapped to a probabilistic form; θ is the time decay factor, which is a learnable parameter that represents the decay effect of authority over time and can be defined as θ = exp(-λt), λ is the decay rate, and t is time.

[0049] In the above method, the constructed three-dimensional tensor plays the role of a core data processing and coordination hub in the university document management system. By constructing a three-dimensional tensor state space and combining the flow probability function with dynamic permission control, the efficiency and security of document management are significantly improved. Compared with the traditional method, the present invention optimizes the flow path through feature vector mapping and probability calculation, and performs permission control based on the flow path. The permission control adopts multi-dimensional feature matching and time decay mechanism to reduce permission conflict events, improve processing efficiency, and ensure that access rights are highly matched with status requirements.

[0050] When the initial state tensor is generated in step S101, the document flow probability α ijk Calculated by Lévy flight random process, the specific formula is:

[0051]

[0052] Where: μ∈(1,2] is the stability index; r is the number of process steps; Γ(1+μ) is the value of the gamma function at (1+μ), which is used to standardize the Lévy distribution;

[0053] When the value of the time sensitivity parameter Γ exceeds the preset threshold, the value of r is increased to recalculate the corresponding α ijk , and according to α ijk The non-zero characteristic of triggers long-distance state jumps and reduces intermediate links. The closer μ is to 1, the heavier the tail of the distribution, the higher the probability of a large jump distance, and the more inclined to long-distance jumps; the closer μ is to 2, the closer the distribution is to the normal distribution, the higher the probability of a small jump distance, and the more inclined to local movement.

[0054] α ijk As a component of the three-dimensional tensor S, it indirectly supports the calculation of the flow probability Access(s1,u) by affecting the generation of Φ(s) and Ψ(s′).

[0055] Through the above random process calculation formula, when r is small, the α value is high, allowing documents to flow along the standard path. When the time sensitivity T exceeds the preset threshold, the Lévy jump mechanism is triggered to directly adjust the document status, which can regulate the burstiness of document flow and adapt to the non-uniformity of tasks in university administration.

[0056] The dynamic weight of the authority γ ijk Updated by the hypergraph neural network, the update formula is:

[0057] γ t+1 =GAT(γ t ,A);

[0058] Where: γ t is the authority weight at the current time t; γ t+1 is the updated authority weight at the next time (t+1); GAT is the graph attention layer; A is the hypergraph adjacency matrix, which represents the multidimensional relationship between departments, users, and documents.

[0059] In the above hypergraph neural network processing, the input data is processed through the graph attention layer (GAT), the dependency of permissions is captured, and γ is updated. ijk For example, in the approval process of scientific research contracts, in the approval stage (j = 2), γ ijk Automatically associate the permissions of the Research Department and the Finance Department to improve the access capabilities of related users. Combined with the time decay factor θ=exp(-λt), when the time t exceeds the preset threshold, γ ijk Reduce and automatically revoke access rights.

[0060] In the above process, the hypergraph neural network is used to update γ in real time. ijk , so that permission control adapts to process status and time changes, which is better than the static permission model; time decay and dependency capture mechanisms ensure that permissions are only valid when necessary, reduce permission conflicts and abuse risks, automatically associate related permissions, reduce manual configuration, and improve administrative efficiency.

[0061] The departmental synergy coefficient β in the three-dimensional tensor ijk When it is lower than the preset threshold, the quantum annealing machine is called to solve the Hamiltonian H to generate the optimal department combination; wherein the Hamiltonian H is solved based on the following formula:

[0062]

[0063]

[0064] in, represents the contribution of the ath department in the collaborative combination v, v is the utility function of the department set, and measures the collaborative effect; A is the set of all departments; Q is the department subset, and |Q| is the number of departments in the subset; is the Paula-Z operator of the ith sector, When the value of is +1 or -1, it means that the a-th department is selected to participate in the collaboration or not; represents the Paula-X operator between the a-th sector and the b-th sector, Optimize the collaboration between departments; J is the intensity of collaboration.

[0065] The above method can dynamically adapt to changes in administrative processes, demonstrate high efficiency and flexibility, and provide an efficient optimization solution for university office document management.

[0066] The following is a specific implementation example: Assume that a university office needs to handle the approval of a scientific research contract, involving the scientific research department, the financial department, and the personnel department. The process starts with the construction of the state space. The system first generates a three-dimensional tensor to represent the state space of document management by inputting the scientific research contract template, the approval process specification, and the time constraint table. This tensor contains the document type (scientific research contract), process status (drafting, preliminary review, approval, archiving, etc.) and time sensitivity (regular, urgent, overdue). Each state point records the flow probability, department coordination coefficient, and authority weight. For example, when the scientific research contract is in the drafting stage and the processing time is regular, the flow probability is high, and it supports the flow along the standard path. Next, the system obtains the electronic spreadsheet of this scientific research contract, extracts its feature vector, and locates it to the coordinates of the drafting state. When the scientific research contract is just drafted, the system calculates the probability of it flowing to preliminary review or approval. Based on the feature difference between the current state and the target state, the preliminary review path is preferred. However, on the sixth day, the remaining time was less than 24 hours, and the time sensitivity became urgent. After the system detected this change, it increased the process step, triggered the jump mechanism, and directly adjusted the document to the approval state, skipping the preliminary review stage. Subsequently, the system determined that the next state was preliminary review and calculated the user permissions in real time. For example, the staff of the preliminary review department tried to access the approved contract, but because the permission requirements in the approval stage were reduced and the time decay effect was not significant, their permission value was lower than the threshold and they could not access it. At the same time, the collaborative efficiency of the preliminary review stage on the previous day dropped to less than the threshold. The system called the quantum annealing machine, and by evaluating the contributions and collaborative relationships of each department, optimized the combination into the Research Department and the Finance Department, excluding the participation of the Personnel Department. This entire process dynamically tracks the state through tensor modeling, combines probability calculation to adjust the path, uses hypergraph neural network to update the permission weight, and uses quantum optimization to improve collaboration efficiency, ultimately reducing the time from 7 days to 5 days, demonstrating high efficiency and flexibility.

[0067] The present invention provides an automated management system for administrative documents in university offices. Figure 2 As shown, the system comprises:

[0068] Construction module 210, which is used to construct a three-dimensional tensor As the state space of document management, dimension N is the document type space, dimension M is the process state space, and dimension T is the time sensitivity parameter space; the initial state tensor is generated by inputting document template, process specification and time constraint table where α ijk is the document flow probability, β ijk is the department synergy coefficient, γ ijk is the dynamic weight of the permission; the subscripts i, j, and k correspond to the specific indexes of the three dimensions in the state space respectively;

[0069] Extraction module 220, which is used to obtain the document to be processed and extract the feature vector Map the document to the corresponding coordinates (i, j, k) in the state space; where s represents the current system state, Φ(s) represents the feature mapping of s in the state space, s′ represents the target state, and Ψ(s′) represents the feature prediction of s′ in the state space;

[0070] The calculation module 230 calculates the flow probability of the document from the current state s to the target state s′ based on the flow probability function, and triggers the Lévy jump mechanism when the time sensitivity parameter T exceeds the threshold, thereby adjusting the flow path of the document; wherein the flow probability function is:

[0071]

[0072] Where: P(s→s′) is the probability of going from the current system state to the target state s′; Z is the normalization factor, and W is the weight matrix generated by adversarial training;

[0073] The permission control module 240 is used to determine the next state s1 based on the flow path, perform dynamic permission control, and calculate the access rights of each user to the state s1 in real time:

[0074]

[0075] Among them, Access(s1,u) represents the quantitative value of whether user u has access rights in document state s1; u represents the vector or identifier of user attributes; K is the total number of permission features; π k (s1) is the k-th level authority characteristic function, which is related to γ ijk Related; k is the original value of user u on the kth feature, will u k Mapped to a probability form; θ is the time decay factor.

[0076] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principle of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A method for automated management of administrative documents in university offices, characterized in that: The method comprises: Constructing a 3D tensor As the state space of document management, dimension N is the document type space, dimension M is the process state space, and dimension T is the time sensitivity parameter space; the initial state tensor is generated by inputting document template, process specification and time constraint table where α ijk is the document circulation probability, β ijk is the department synergy coefficient, γ ijk is the dynamic weight of the permission; the subscripts i, j, and k correspond to the specific indexes of the three dimensions in the state space respectively; Obtain the document to be processed, extract the feature vector V = Φ(s) ⊕ Ψ(s′), and map the document to the corresponding coordinates (i, j, k) in the state space; wherein s represents the current system state, Φ(s) represents the feature mapping of s in the state space, s′ represents the target state, and Ψ(s′) represents the feature prediction of s′ in the state space; The flow probability of the document from the current state s to the target state s′ is calculated based on the flow probability function, and when the time sensitivity parameter T exceeds the threshold, the Lévy jump mechanism is triggered to adjust the flow path of the document accordingly; wherein the flow probability function is: Where: P(s→s′) is the probability of going from the current system state to the target state s′; Z is the normalization factor, and W is the weight matrix generated by adversarial training; Based on the flow path, the next state s1 is determined, dynamic permission control is performed, and the access rights of each user to state s1 are calculated in real time: Among them, Access(s1,u) represents the quantitative value of whether user u has access rights in document state s1; u represents the vector or identifier of user attributes; K is the total number of permission features; π k (s1) is the k-th level authority characteristic function, which is related to γ ijk Related; k is the original value of user u on the kth feature, will u k Mapped to a probability form; θ is the time decay factor.

2. The automated management method for administrative documents in university offices according to claim 1 is characterized in that: When generating the initial state tensor, the document flow probability α ijk Calculated by Lévy flight random process, the specific formula is: Where: μ∈(1,2] is the stability index; r is the number of process steps; Γ(1+μ) is the value of the gamma function at (1+μ), which is used to standardize the Lévy distribution; When the value of the time sensitivity parameter T exceeds the preset threshold, the value of r is increased to recalculate the corresponding α ijk , and according to α ijk The non-zero characteristic of triggers long-distance state jumps, reducing intermediate links.

3. The automated management method for administrative documents in university offices according to claim 1 is characterized in that: The dynamic weight of the authority γ ijk Updated by the hypergraph neural network, the update formula is: c t+1 =GAT(γ t ,A); Where: γ t is the authority weight at the current time t; γ t+1 is the updated authority weight at the next time (t+1); GAT is the graph attention layer; A is the hypergraph adjacency matrix, which represents the multidimensional relationship between departments, users, and documents.

4. The automated management method for administrative documents in university offices according to claim 1 is characterized in that: The departmental synergy coefficient β in the three-dimensional tensor ijk When it is lower than the preset threshold, the quantum annealing machine is called to solve the Hamiltonian H to generate the optimal department combination; wherein the Hamiltonian H is solved based on the following formula: in, represents the contribution of the ath department in the collaborative combination v, v is the utility function of the department set, and measures the collaborative effect; A is the set of all departments; Q is the department subset, and |Q| is the number of departments in the subset; is the Paula-Z operator of the ith department; represents the Paula-X operator between the a-th department and the b-th department; J is the collaboration intensity.

5. An automated management system for administrative documents in university offices, characterized in that: The system comprises: Building blocks for constructing three-dimensional tensors As the state space of document management, dimension N is the document type space, dimension M is the process state space, and dimension T is the time sensitivity parameter space; the initial state tensor is generated by inputting document template, process specification and time constraint table where α ijk is the document circulation probability, β ijk is the department synergy coefficient, γ ijk is the dynamic weight of the permission; the subscripts i, j, and k correspond to the specific indexes of the three dimensions in the state space respectively; An extraction module, which is used to obtain a document to be processed, extract a feature vector V = Φ(s) ⊕ Ψ(s′), and map the document to the corresponding coordinates (i, j, k) in the state space; wherein s represents the current system state, Φ(s) represents the feature mapping of s in the state space, s′ represents the target state, and Ψ(s′) represents the feature prediction of s′ in the state space; The calculation module calculates the flow probability of the document from the current state s to the target state s′ based on the flow probability function, and when the time sensitivity parameter T exceeds the threshold, triggers the Lévy jump mechanism to adjust the flow path of the document accordingly; wherein the flow probability function is: Where: P(s→s′) is the probability of going from the current system state to the target state s′; Z is the normalization factor, and W is the weight matrix generated by adversarial training; The permission control module is used to determine the next state s1 based on the flow path, perform dynamic permission control, and calculate the access rights of each user to the state s1 in real time: Among them, Access(s1,u) represents the quantitative value of whether user u has access rights in document state s1; u represents the vector or identifier of user attributes; K is the total number of permission features; π k (s1) is the k-th level authority characteristic function, which is related to γ ijk Related; k is the original value of user u on the kth feature, will u k Mapped to a probability form; θ is the time decay factor.

Images