JP2005070919A - Recall type information storage device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a recall type information device for extracting useful related information from existing information in response to a user's key information input.
A database 103 storing existing information and a degree of recall, input of parameters regarding how to extract related information, input of key information, input of information newly registered as existing information, and extracted related information The input / output means 101 for outputting, the similarity calculation means 104 for calculating the similarity between the key information and the existing information, and the recall of each existing information stored in the database 103, the key information and the existing The degree of recall given to each existing information is calculated using the similarity between information, the degree of similarity between existing information, and parameters, and is set in advance from the database 103 and the recall degree calculating means 105 that changes over time. The control means 102 which takes out the existing information to which the recall degree more than the threshold value was given as the relevant information of key information is provided.
[Selection] Figure 1

Description

  The present invention relates to a technology for supporting storage and retrieval of information in a personal information terminal or the like, and in particular, a recollective information storage that retrieves useful related information from existing information held by a user in response to input of key information by the user. It relates to the device.

  With the spread of personal information terminals such as computer networks and PDAs (Personal Digital Assistants), a large amount of information can be easily collected. Since there is a variety of information such as e-mails, memos, WWW content, and photos, this information cannot be flexibly handled by a method that is strictly classified and managed hierarchically according to the content and purpose.

  Conventionally, for example, the following methods have been proposed as an organization and management method for efficiently storing such a large amount of diverse information and efficiently extracting only useful information as necessary. Yes. The first method is a method in which information is arranged and managed in chronological order regardless of the contents (see, for example, Non-Patent Document 1), and avoids problems such as setting of classification items and in which classification item information should be classified. can do. The second method is a method for time management of the state of the desktop environment of the computer (see, for example, Non-Patent Document 2). Since the arrangement status of information on the desktop is recorded in time order as it is, Work content can be reproduced as needed. In the second method, the change history applied to the desktop environment such as deletion of information and generation of new information is also managed in time order. In particular, when deleting information, the information itself is not completely deleted from the computer, but a work history of “deleted from the desktop” is recorded. Can be traced back to when it was on the desktop. In order to switch the desktop environment along the time axis, there is also provided a function for searching when specified information appears on the desktop, and when it is created / deleted.

  However, the above prior art has the following problems. When users retrieve past information, they may find similar experiences and events in the past for problems and issues they are currently facing, and use them to solve current problems. . At this time, information that matches exactly the same conditions as the current problem does not always exist in the past, so there is a possibility that you can find clues to solve the problem by sequentially referencing related information that satisfies some conditions. is there.

  In the above prior art, when the desired information is clear, it can be extracted retroactively to the time when the information exists, or information existing at the same time as the desired information can be extracted as related information. When there are many time points to be processed or when there is little related information included in each time point, this work is burdensome for the user. In addition, there are various types of information depending on the purpose and activity at each point in time, and it may not be said that there is a useful relationship between these pieces of information just because they existed at the same point in time. It is not effective to extract relevant information back in time. Furthermore, if too much information is extracted as related information, truly useful information is buried, so a means for controlling the number of related information to be extracted is required.

E. Freeman and D.C. Gelernter, Lifestreams: “A storage model for personal data”, ACM SIGMOD Bulletin, 28, March 1996. Jun Rekimoto, Time-Machine Computing: "A Time-centric Approach for the Information Environment", ACMIST'99, 1999.

The present invention has been made to solve the above problems, and is directly related to key information based on similarity between existing information and information used as a key for retrieving related information and existing information. Another object of the present invention is to provide a recall type information storage device that can extract not only existing information but also indirectly related existing information as related information.
In addition, according to the process in which the related information has been extracted, the present invention relates only to useful existing information by adjusting the recall level indicating whether or not the information given to the existing information should be extracted. An object of the present invention is to provide a recall type information storage device that can be taken out as information.
In addition, the present invention adjusts the number of related information to be extracted by controlling the average value of the degree of recall, and adjusts the influence of the process in which the related information has been extracted on the extraction of new related information. It is an object of the present invention to provide a recall type information storage device that enables the above.

  The recall type information storage device of the present invention includes a database storing existing information and a degree of recall indicating whether or not the existing information should be extracted, input of parameters regarding how to extract related information, input of key information, Input / output means for inputting information to be newly registered as existing information and outputting the extracted related information, and similarity calculation means for calculating similarity between key information and existing information and similarity between existing information And using parameters relating to the recall of each existing information stored in the database, the similarity between the key information and the existing information, the similarity between the existing information, and how to extract the related information, The degree of recall given to the existing information is calculated and the degree of recall is calculated, and the existing information to which the degree of recall equal to or higher than a preset threshold is given from the database. And a control means for retrieving the related information of the key information.

  Here, existing information is used as an episode, and information used as a key for extracting a related episode from the episode is referred to as a key episode. In addition, for key episodes and episodes, attribute information (hereinafter referred to as meta information) for explaining itself is also provided.

  Similarities indicating key episodes and similarities between episodes are calculated using meta information. The episode is given a recall level indicating whether or not it should be taken out, and when the value exceeds a set threshold, the episode is presented to the user as a related episode.

  For each episode, if a key episode similar to itself is entered, and if there are other episodes that have a high degree of recollection and high relevance, the degree of recollection Increase with time in proportion to product. Thereby, the episode directly related to the key episode can be taken out, and the episode related to the directly related episode can also be taken out indirectly as the episode related to the key episode.

  On the other hand, with respect to other similar episodes, each episode decreases the opponent's recall level with time in proportion to the product of the similarity level and the recall level. This suppressive action of the recall level prevents the recall level from increasing in a number of episodes. In addition to adjusting the recall level between these episodes, the recall level of each episode is decreased with time in proportion to its size, so that no other related episodes exist or related key episodes are entered For episodes that are not performed, the degree of recall decreases with time as an unneeded old episode, making it impossible to extract.

  In the calculation of the degree of recall, the degree of recall increases as the number of related key episodes input, and the degree of recall changes with time depending on the current degree of recall. Therefore, when the relevance is determined based on the similarity only, the same episode is always taken out as a related episode, but in the present invention, the recollection degree changes depending on the history of the key episode being input, and more in the past. The episodes that are pulled out are extracted as useful episodes.

  The number of related episodes and the influence of the input history of key episodes are adjusted by controlling the average recall level at each time point so as to approach a preset target average value. When a large target average value is set, the recall level is allowed to increase in each episode, so episodes exceeding the threshold increase and many related episodes are extracted. In addition, the recall of episodes related to past key episodes is allowed to remain large even if they are not related to the current key episode. growing. On the other hand, if the target average value is small, only the episode related to the current key episode is allowed to be recalled, so the number of episodes exceeding the threshold decreases, and the input history of key episodes up to that point is reduced. The impact is also reduced.

  According to the present invention, it is possible to extract not only episodes directly related to key episodes but also indirectly related episodes as related episodes. Further, only useful episodes can be extracted as related episodes according to the process of extracting related episodes. Furthermore, it is possible to adjust the number of related episodes and to adjust the influence of the process in which related episodes have been extracted on the extraction of new related episodes.

  Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings.

  FIG. 1 shows an overall configuration diagram of an embodiment of a recall type information storage apparatus according to the present invention. In FIG. 1, 101 is an input / output unit, 102 is a control unit, 103 is an episode database, 104 is a similarity calculation unit, and 105 is a recall level calculation unit. Actually, the recall type information storage device is constituted by a computer.

  The input / output unit 101 inputs information (key episode) that is a key for extracting related information from existing information (episode), inputs parameters related to how related information is extracted, and information to be newly registered as existing information (episode ) And display of the extracted related information. The control unit 102 controls input / output operations of information and other operations of the entire apparatus. The episode database 103 stores existing information (episode), attribute information (meta information), and a recall level indicating whether or not the existing information should be extracted. In the embodiment, the episode database 103 stores the similarity between the existing information calculated by the similarity calculation unit 104, and further stores the similarity between the key information and the existing information temporarily. The similarity calculation unit 104 calculates the similarity between key information (key episode) and existing information (episode), and the similarity between existing information (episode). The recall level calculation unit 105 calculates the recall level given to each existing information (key episode) in the episode database 103 and indicating the degree of whether or not the existing information should be extracted, and changes the time.

  The outline of the operation of this embodiment will be described below. In the following description, it is assumed that existing information is an episode, and information serving as a key for extracting a related episode from the episode is a key episode.

  FIG. 2 is a flowchart showing an outline of the operation when an episode to be newly registered is input to the episode database 103. The user uses the input / output unit 101 to input a new episode and an initial value of the recall level (step 201). The control unit 102 newly registers the input episode and the initial value of the recall level in the episode database 103 (step 202). Next, the control unit 102 retrieves an existing episode from the episode database 103 and outputs it to the similarity calculation unit 104 together with the newly input episode. The similarity calculation unit 104 uses the meta information of the episode as a basis. The control unit 102 receives the calculation result of the similarity between episodes from the similarity calculation unit 104 and registers it in the episode database 103 (step 203). Since the similarity calculation itself is not a feature of the present invention, its details are omitted. Next, the control unit 102 extracts the recall level of episodes and the similarity between episodes from the episode database 103 and outputs them to the recall level calculation unit 105. The recall level calculation unit 105 calculates the recall level of episodes and performs control. The unit 102 receives the calculation result of the recall level from the recall level calculation unit 105, and registers (updates) it again in the episode database 103 (step 204). The processing of the recall level calculation unit 105 will be described later. Further, it is assumed that parameters input from the input / output unit 101 are set in the recall level calculation unit 105 in advance via the control unit 102.

  FIG. 3 is a flowchart illustrating an operation outline when a key episode is input. The user inputs a key episode using the input / output unit 101 (step 301). The control unit 102 temporarily holds the input key episode. Next, the control unit 102 retrieves the episode from the episode database 103 and outputs it to the similarity calculation unit 104 together with the input key episode, and the similarity calculation unit 104 calculates the similarity between the key episode and the episode. Then, the control unit 102 receives the calculation result of the similarity between the key episode and the episode from the similarity calculation unit 104, and temporarily registers it in the episode database 103 (step 302). Next, the control unit 102 retrieves the recall level of the episode, the similarity between episodes, and the similarity between the key episode and the episode from the episode database 103 and outputs them to the recall level calculation unit 105. The degree of recall of the episode is calculated, and the control unit 102 receives the calculation result of the degree of recall from the recall level calculation unit 105 and registers (updates) it again in the episode database 103 (step 303). Next, the control unit 102 scans the episode database 103, reads episodes having a degree of recall equal to or higher than a preset threshold value, and outputs them to the input / output unit 101, so that episodes related to the key episode are retrieved. Present to the user (step 304). Finally, the control unit 102 clears the input key episode, and the similarity between the key episode and the episode from the episode database 103 so as not to affect the recall calculation with the next input key episode. Is deleted (step 305).

ここで、ｘ_ｊはエピソードｊの想起度、ｍ_ｊｉはエピソードｉから見たエピソードｊの類似度、ｍ_ｉｊはエピソードｊから見たエピソードｉの類似度であり、ｍ_ｉはエピソードｉから見たキーエピソードの類似度である。ｍ_ｊｉ及びｍ_ｉｊは一度計算されると、それ以降はエピソードデータベース１０３に保持される。ｍ_ｊはキーエピソードが入力されたときだけ、一時的にエピソードデータベース１０３に保持され、不要になると削除される（先のステップ３０５）。また、パラメータｃとｋ_２の値、及び、ｋ_１の初期値は、あらかじめユーザが入出力部１０１を用いて入力し、制御部１０２により想起度計算部１０５に保持される。 Next, the process of the recall calculator 105 will be described in detail. The recall level calculator 105 changes the recall level with time according to the following equation, where x _i is the recall level of episode i at a certain time t.

Here, x _j is recalled degree episode j, m _ji similarity episode j viewed from the episode i, m _ij is the similarity episode i viewed from the episode j, m _i is viewed from episodes i It is the similarity of key episodes. Once m _ji and m _ij are calculated, they are stored in the episode database 103 thereafter. m _j is temporarily stored in the episode database 103 only when a key episode is input, and is deleted when it becomes unnecessary (previous step 305). Further, the values of the parameters c and k ₂ and the initial value of k ₁ are previously input by the user using the input / output unit 101, and are stored in the recall level calculation unit 105 by the control unit 102.

The first term on the right-hand side of equation (1) is indirectly related to the key episode by increasing the recall level of episode i when there are other similar episodes with a high recall level as seen from episode i. Work to draw related episodes. The second term works to prevent other episodes that have considered episode i similar to reduce the recall of episode i and prevent the recall from becoming too large in many episodes, k ₁ is a parameter (suppression parameter) related to the strength of the effect. The first and second terms are calculated between episodes where m _ji and m _ij each have a positive value. The third term works to increase the recall level of episode i when a similar key episode as viewed from episode i is input. c is a parameter relating to the speed of temporal change of the recall level by these terms, and the larger the value, the faster the change of the recall level by these terms. Item 4 works to reduce the recall of episode i over time, so if there is no episode similar to episode i or no similar key episode is entered, episode i is not needed old It is regarded as a serious episode and is no longer taken out. k ₂ is a parameter related to the speed of recall of the time variation of the fourth paragraph recall index larger this value decreases rapidly. In the embodiment, by setting the k ₂ to a value smaller than c, gradually recall degree in the absence of other episodes and key episodes so as decreases.

ここで、Ｎはエピソードの数を表す。 Next, control of the average value of recall will be described. The average value of the recall level at each time is brought close to a preset target average value (hereinafter referred to as θ) by adjusting the suppression parameter k _{1 in the} equation (1). The value of k ₁ at time t + 1 (represented as k ₁ [t + 1]) is the document “Kantaris,
N. , Howden, P .; F. , The Universal Equation Solver by BASIC, SIGMA PRESS, 1983 (Mitsui translation, numerical solution of equations, Keigaku Shuppan, 1987).

Here, N represents the number of episodes.

Figure 4 is a flowchart showing an outline of update processing for suppressing the parameter k _1. The value obtained by subtracting θ from the average value of recall at time t (referred to as the recall average difference) is negative, and the increase rate of the average recall is negative, and the recall average difference is positive, and the increase rate of the average value of the recall of the case of a positive, updating the _{k 1} (step 401, step 402 is Yes, step 403 is Yes). In these cases, if the sign of the recall level average difference at time t is the same as the sign of the recall level average difference at time t−1, p in equation (3) is incremented by 1 to accelerate convergence to θ ( Step 404 is Yes, Step 405). Conversely, code increases 1 to r if different, to prevent divergence from the theta (step 404 No, step 406) Then, according to (3), updates the _{k 1} (step 407) The value of the recall average difference is recorded for update at time t + 2 (step 408).

  Next, a specific example of the recall process in the recall calculator 105 will be described. 5 and 6 show that 100 lattice points are randomly arranged as episodes on a 30 × 30 lattice, and the value obtained by dividing the distance between lattice points by the maximum distance is the degree of recall between episodes. The grid points with the larger recall value are displayed darker. 5 shows a case where the target average value θ = 0.2 is set, and FIG. 6 shows a case where θ = 0.4 is set.

5 and 6, the recall level of each grid point is set at random at time t = 0, a key episode is input at the position indicated by A at time t = 200, and A at time t = 400. The key episode input at the position is deleted, and the key episode is input at the position indicated by B in the figure at time t = 600. For the parameters in the equation (1), c = 1, k ₂ = 0.1, and the initial value of k ₁ is set to 0.5. As can be seen at t = 200, the larger the θ, the more lattice points with a large recall. Therefore, it is possible to adjust the number of episodes that exceed a preset threshold of recall by adjusting θ, that is, the number of related episodes. At t = 400, the degree of recall is large at the lattice point near A, indicating that the degree of recall of the related episode is increased by the input of the key episode. Even when the key episode is deleted, the episode related to the previously entered key episode is useful as long as there is no new key episode input, so that the degree of recall hardly changes at t = 600. Is remembered as At t = 800, the degree of recall is large at the grid point near B due to the key episode input at t = 600, but when θ = 0.4, the effect of the key episode previously input is near A. There is little change in recall at the grid points. This is different from the situation where a high-recall grid point exists only in the vicinity of A when a key episode is input for the first time as seen at t = 400. It shows that it depends on the input history. However, when θ = 0.2, the recollection degree of the lattice points near A is almost small, and is not influenced by the input history of the key episode. Therefore, by adjusting θ, it is possible to adjust the influence of the input history of the key episode.

7 and 8 show the results of average value control of the recall level. FIG. 7 shows the case of θ = 0.2, and FIG. 8 shows the case of θ = 0.4. Each thin line indicates a change in the recall level at 100 lattice points (episodes), and each thick line indicates a change in the average value of these recall levels. 7 and 8, by adjusting the inhibition parameters k _1, shows that it is possible to converge the average value of the recall of the target setpoint.

  The recall type information storage device of the present invention has been specifically described based on one embodiment. However, the present invention is not limited to the above-described embodiment, and various modifications can be made without departing from the scope of the patent claims. Needless to say, it can be changed. For example, in the embodiment, the similarity between episodes is stored in the episode database once calculated, but may be calculated each time a key episode or episode is input.

It is a figure which shows the whole structure of one Example of the recalling-type information storage device of this invention. It is a flowchart which shows the outline | summary of the operation | movement at the time of the episode input of one Example of this invention. It is a flowchart which shows the operation | movement outline | summary at the time of the key episode input of one Example of this invention. It is a flowchart which shows the operation | movement outline | summary of the average control of recall. It is a figure which shows the specific example (in the case of (theta) = 0.2) of the process of recall. It is a figure which shows the specific example (in the case of (theta) = 0.4) of the process of recall. It is a figure which shows the operation example (in the case of (theta) = 0.2) of average value control of recall. It is a figure which shows the operation example (when (theta) = 0.4) of average value control of recall.

Explanation of symbols

101 I / O Unit 102 Control Unit 103 Episode Database 104 Similarity Calculation Unit 105 Recall Level Calculation Unit

Claims (5)

A recall type information storage device that extracts useful related information from existing information in response to input of key information (hereinafter, key information) for extracting related information from existing information,
A database storing existing information and a degree of recall indicating whether or not the existing information should be extracted;
Input / output means for inputting parameters relating to how to extract related information, inputting key information, inputting information newly registered as existing information, and outputting extracted related information;
Similarity calculation means for calculating the similarity between key information and existing information and the similarity between existing information;
Each existing information is stored using parameters relating to the recall level of each existing information stored in the database, the similarity between the key information and the existing information, the similarity between the existing information, and how to extract the related information. Recall degree calculation means for calculating the degree of recall given to
From the database, a control means for extracting existing information provided with a recall level equal to or higher than a preset threshold as related information of key information;
A recall type information storage device comprising: The recall type information storage device according to claim 1,
The recall level calculation means uses the following formula as the recall level of the existing information i at a certain time t.

x _j is the recall of existing information j at time t m _ji is the similarity of existing information j viewed from existing information i m _ij is the similarity of existing information i viewed from existing information j mi is viewed from existing information i The key information similarity c, k ₁ , k ₂ is a recall type information storage device characterized in that the recall degree is changed over time according to a parameter relating to how relevant information is extracted. The recall type information storage device according to claim 2,
The recall type information storage device, wherein the parameter k2 in the equation (1) is set to a value smaller than the parameter c. The recall type information storage device according to claim 2 or 3,
The recall level information storage unit is configured to update the parameter k1 so as to control the average value of the recall level of each existing information at each time to approach a preset target average value. apparatus. The recall type information storage device according to claim 4,
The recall level calculation means has a negative value (hereinafter referred to as recall level average difference) obtained by subtracting the target average value from the average value of recall level at time t, and the increase rate of the average value of recall level is negative, and The recall type information storage device, wherein the parameter k1 is updated when the average value of the recall level is positive and the increase rate of the average value of the recall level is positive.

Images